Wind farm power generation prediction method and device based on graph attention mechanism

Abstract

The present application proposes a method and device for predicting wind farm power generation based on a graph attention mechanism, which relates to the field of wind power generation technology. The method includes: determining the current turbine and the upstream neighbors of the current turbine; determining a pre-constructed wind farm network directed graph, and obtaining characteristic information of the current turbine based on the network directed graph and the upstream neighbors; obtaining a first node representation of the current turbine and a second node representation of the upstream neighbors based on the distance feature and the angle feature; obtaining a third node representation of the current turbine based on the first node representation, the second node representation, and a preset graph attention GAT network model; and obtaining the predicted power of the current turbine based on the third node representation and a preset fully connected neural network MLP model. The present application utilizes a graph attention mechanism to assign different weights to upstream neighbors that have an impact on the current turbine, predict the power of the current turbine, and to a certain extent solve the problem of inaccurate power prediction caused by the wake effect.

Description

Technical Field

[0001] The present application relates to the field of wind power generation technology, and in particular to a method and device for predicting wind farm power generation based on a graph attention mechanism. Background Art

[0002] Wind power generation has developed rapidly in recent years. As the penetration of wind energy in the power grid increases, predicting the expected power generation of large wind farms becomes increasingly important.

[0003] Related technologies often use turbine power curves to predict power generation at different wind speeds. However, with the construction of larger wind farms, turbines located close together can interact with each other, causing actual production to deviate from the power prediction curves. Insufficient wind speed and increased turbulence downstream of the turbines reduce the power generated by downstream turbines. Therefore, accounting for these wake losses and modeling the complex flow patterns generated downstream of wind farms to accurately predict wind farm power generation have become key research areas in this field. Summary of the Invention

[0004] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] To this end, the first aspect of the present application proposes a wind farm power prediction method based on a graph attention mechanism, which is characterized by comprising the following steps:

[0006] Determining a current turbine and an upstream neighbor of the current turbine; wherein the upstream neighbor is a turbine among upstream turbines of the current turbine that meets a preset condition, wherein the preset condition is that an angle between a line connecting the upstream turbine and the current turbine and a wind direction of the wind farm is less than a preset threshold;

[0007] Determining a pre-constructed wind farm network directed graph, and obtaining characteristic information of the current turbine based on the network directed graph and the upstream neighbor; wherein the characteristic information of the current turbine includes a distance characteristic between the current turbine and the upstream neighbor, and an angle characteristic between the current turbine and the upstream neighbor;

[0008] Obtaining a first node representation of the current turbine and a second node representation of the upstream neighbor according to the distance feature and the angle feature;

[0009] Obtaining a third node representation of the current turbine based on the first node representation, the second node representation, and a preset graph attention GAT network model; wherein the GAT network model has learned the ability to assign different weights to the current turbine and its different neighboring turbines;

[0010] The predicted power of the current turbine is obtained according to the third node representation and a preset fully connected neural network MLP model; wherein the MLP model has learned the ability to predict turbine power.

[0011] In some embodiments of the present application, the third node representation of the current turbine is obtained based on the first node representation, the second node representation and the graph attention network GAT model, including: updating the first node representation through the graph attention network GAT model to obtain a fourth node representation of the current turbine; obtaining the attention coefficient between the current turbine and the upstream neighbor based on the fourth node representation and the second node representation; and obtaining the third node representation of the current turbine by weighted summation of the attention coefficient between the current turbine and the upstream neighbor, the fourth node representation and the second node representation.

[0012] In some embodiments of the present application, the GAT graph attention network model updates the first node representation using the following formula to obtain a fourth node representation of the current turbine:

[0013] h′ (i,j) =f edge (h i ,h j ,x (i,j) )

[0014]

[0015] Where h′ (i,j) represents the updated representation of the edge between the current turbine i and the upstream neighbor j in the neural network, h i The first node representation of the current turbine i, h j is the second node representation of the upstream neighbor j, h′ i is the fourth node representation of the current turbine.

[0016] In some embodiments of the present application, obtaining the attention coefficient between the current turbine and the upstream neighbor of the current turbine based on the fourth node representation and the second node representation includes: obtaining the similarity coefficient between the current turbine and the upstream neighbor of the current turbine based on the fourth node representation and the second node representation; normalizing the similarity coefficient between the current turbine and the upstream neighbor of the current turbine by a softmax function to obtain the attention coefficient between the current turbine and the upstream neighbor of the current turbine.

[0017] In some embodiments of the present application, the wind farm network directed graph is pre-constructed in the following manner: a network directed graph of the wind farm is constructed based on the wind farm conditions and the wind farm layout; wherein the nodes in the network directed graph represent the turbines in the wind farm, and there is an edge between the node corresponding to each of the turbines and the node corresponding to the upstream neighbor of each of the turbines, and the direction of the edge is from the upstream neighbor of each of the turbines to the corresponding turbine.

[0018] In some embodiments of the present application, obtaining the predicted power of the current turbine based on the third node representation and the preset fully connected neural network MLP model includes: determining the airflow speed of the wind farm; obtaining the predicted power of the current turbine based on the third node representation, the airflow speed of the wind farm and the preset fully connected neural network MLP model.

[0019] In a second aspect, the present application proposes a wind farm power prediction device based on a graph attention mechanism, which is characterized by comprising:

[0020] a determination module, configured to determine a current turbine and an upstream neighbor of the current turbine; wherein the upstream neighbor is a turbine among upstream turbines of the current turbine that satisfies a preset condition, wherein the preset condition is that an angle between a line connecting the upstream turbine and the current turbine and a wind direction of the wind farm is less than a preset threshold;

[0021] a first acquisition module, configured to determine a pre-constructed wind farm network directed graph, and acquire characteristic information of the current turbine based on the network directed graph and the upstream neighbor; wherein the characteristic information of the current turbine includes a distance characteristic between the current turbine and the upstream neighbor, and an angle characteristic between the current turbine and the upstream neighbor;

[0022] a second acquisition module, configured to acquire a first node representation of the current turbine and a second node representation of the upstream neighbor according to the distance feature and the angle feature;

[0023] a third acquisition module, configured to obtain a third node representation of the current turbine based on the first node representation, the second node representation, and a preset graph attention (GAT) network model; wherein the GAT network model has learned to assign different weights to the current turbine and its different neighboring turbines;

[0024] A fourth acquisition module is used to obtain the predicted power of the current turbine based on the third node representation and a preset fully connected neural network MLP model; wherein the MLP model has learned the ability to predict turbine power.

[0025] In some embodiments of the present application, the third acquisition module is specifically used to: update the first node representation through the graph attention network GAT model to obtain the fourth node representation of the current turbine; obtain the attention coefficient between the current turbine and the upstream neighbor based on the fourth node representation and the second node representation; and obtain the third node representation of the current turbine by weighted summation of the attention coefficient between the current turbine and the upstream neighbor, the fourth node representation and the second node representation.

[0026] In a third aspect, the present application proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the first aspect is implemented.

[0027] In a fourth aspect, the present application proposes a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in the first aspect above.

[0028] The wind farm power prediction method based on the graph attention mechanism in the embodiments of this application utilizes the graph attention mechanism and adopts deep learning methods for modeling. Different weights are assigned to upstream neighbors that influence the current turbine, and the upstream turbine with the greatest influence on the current turbine is determined. This method then obtains the predicted power of the current turbine, improving prediction accuracy while reducing computational costs. This application can, to a certain extent, address the issue of inaccurate wind farm power predictions caused by the wake effect and reduce computational costs.

[0029] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0031] Figure 1 A schematic diagram of a flow chart of a method for predicting wind farm power generation based on a graph attention mechanism provided in an embodiment of the present application;

[0032] Figure 2 A schematic diagram of training a graph attention GAT network model and a preset fully connected neural network MLP model provided in an embodiment of the present application;

[0033] Figure 3A flowchart of a method for obtaining a third node representation of a current turbine through a graph attention network (GAT) model provided in an embodiment of the present application;

[0034] Figure 4 A schematic diagram of an embodiment of the present application providing a method for normalizing similarity coefficients between a current turbine and its upstream neighbors using a softmax function;

[0035] Figure 5 A schematic diagram of a wind farm power prediction device based on a graph attention mechanism provided in an embodiment of the present application;

[0036] Figure 6 This is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0038] In the related art, in order to explain the wake loss in the wind farm, the complex flow pattern generated downstream of the wind farm is modeled. For example, the Jensen wake model calculates the vortex model and the velocity loss downstream of a single wind turbine. Computational fluid dynamics can produce accurate results, but such numerical methods will incur high computational costs. The present application proposes a method and device for predicting wind farm power generation based on a graph attention mechanism. The graph attention mechanism is used to capture the weights of different upstream turbines of the current turbine and predict the power generation of the current turbine. While improving the accuracy of the predicted power generation, it can also reduce some of the computational costs. Specifically, the following describes the method and device for predicting wind farm power generation based on a graph attention mechanism according to an embodiment of the present application with reference to the accompanying drawings.

[0039] Figure 1 This is a flow chart of a method for predicting wind farm power generation based on a graph attention mechanism provided in an embodiment of the present application. Figure 1 As shown in FIG, the wind farm power generation prediction method based on the graph attention mechanism includes the following steps:

[0040] Step 101: Determine the current turbine and its upstream neighbors. The upstream neighbors are turbines upstream of the current turbine that meet a preset condition: the angle between the line connecting the upstream turbine and the current turbine and the wind direction of the wind farm is less than a preset threshold.

[0041] It's important to note that because the wake effect is the influence of upstream turbines on downstream turbines, when predicting the power generation of a current turbine, it's necessary to identify upstream neighbors that affect the current turbine's power generation. As an example, the preset threshold could be 30 degrees. That is, among the upstream turbines of the current turbine, those upstream turbines whose line with the current turbine makes an angle with the wind direction less than 30 degrees are considered upstream neighbors of the current turbine.

[0042] Step 102: Determine a pre-built wind farm network directed graph and obtain characteristic information of the current turbine based on the network directed graph and its upstream neighbors. The characteristic information of the current turbine includes the distance characteristic between the current turbine and its upstream neighbors, and the angle characteristic between the current turbine and its upstream neighbors.

[0043] In some embodiments of the present application, the characteristic information of the current turbine includes the distance characteristic D between the current turbine i and its upstream neighbor j. ij , and the angle characteristic between the current turbine i and its upstream neighbor j where α is the angle between the line connecting the current turbine i and its upstream neighbor j and the wind direction, and t is the current moment.

[0044] In some embodiments of the present application, a network directed graph for a wind farm can be pre-constructed based on wind farm conditions and wind farm layout. Optionally, the wind farm conditions may include wind speed and direction within the wind farm. Nodes in the network directed graph represent turbines within the wind farm. Each turbine node has an edge with its upstream neighbor node. The edge is directed from the upstream neighbor to the turbine, representing the impact of the upstream turbine on the downstream turbine.

[0045] Step 103 : Obtain a first node representation of the current turbine and a second node representation of the upstream neighbor according to the distance feature and the angle feature.

[0046] In some embodiments of the present application, the characteristic information of the current turbine includes the distance characteristic D between the current turbine i and its upstream neighbor j. ij , and the angle characteristic between the current turbine i and its upstream neighbor j Obtain the first node representation h of the current turbine i through a layer of MLP i and the second node of upstream neighbor j represents h j .

[0047] Step 104: Obtain a third node representation of the current turbine based on the first node representation, the second node representation, and a preset graph attention GAT network model. The GAT network model has learned to assign different weights to the current turbine and its neighboring turbines.

[0048] It should be noted that the graph attention GAT network model is a model based on the spatial domain. Among them, the attention mechanism is used to update the node representation (i.e., the characterization vector) of the current turbine in the network directed graph, and assign different attention weights to the current turbine and its upstream neighbors. It should be noted that the attention weight can be regarded as the relative importance of the network to different parts of the input, and the graph attention GAT network model can focus on the most influential upstream turbine in the process of learning to assign weights to determine the wake loss. In some embodiments of the present application, the graph attention GAT network model can obtain a third node representation of the current turbine by calculating the attention coefficient between the current turbine and its upstream neighbors and performing a weighted summation of the features.

[0049] Step 105: Obtain the predicted power of the current turbine based on the third node representation and the preset fully connected neural network MLP model, wherein the MLP model has learned the ability to predict turbine power.

[0050] As a possible implementation method, in some embodiments of the present application, the wind speed ws of the wind farm can be determined. t , and according to the third node h i , wind farm airflow speed ws t With the preset fully connected neural network MLP model, the predicted power of the current turbine is obtained through the following formula.

[0051]

[0052] in, is the predicted power of turbine i at time t.

[0053] In some embodiments of the present application, the generated power of turbines in a wind farm may be predicted separately to obtain the predicted power of each turbine, thereby obtaining a power prediction curve of the wind farm.

[0054] It should be noted that the preset graph attention GAT network model and the preset fully connected neural network MLP model in the embodiments of the present application are both obtained through pre-training. Figure 2As shown, historical data of the wind farm at different times (including historical data such as wind direction at different times, characteristic information of the current turbine, predicted power and actual power) can be obtained, and the historical data of the wind farm can be divided into a training set and a test set according to a ratio of 3:1. The graph attention GAT network model and the fully connected neural network MLP model are trained so that the GAT network model learns to obtain the ability to assign different weights to the current turbine and its different neighboring turbines, so that the MLP model learns to obtain the ability to predict turbine power. Optionally, in some embodiments of the present application, after obtaining the predicted power of the current turbine, the wind direction at the current moment, the characteristic information of the current turbine, the predicted power and the actual power can be used as historical data, and the graph attention GAT network model and the fully connected neural network MLP model are iteratively trained and updated to generate a new network directed graph, and the characteristic information and predicted power of the current turbine are updated at the same time.

[0055] The wind farm power prediction method based on the graph attention mechanism in the embodiments of this application utilizes the graph attention mechanism and adopts deep learning methods for modeling. Different weights are assigned to upstream neighbors that influence the current turbine, and the upstream turbine with the greatest influence on the current turbine is determined. This method then obtains the predicted power of the current turbine, improving prediction accuracy while reducing computational costs. This application can, to a certain extent, address the issue of inaccurate wind farm power predictions caused by the wake effect and reduce computational costs.

[0056] Figure 3 This is a flow chart of a method for obtaining the third node representation of the current turbine through a graph attention network GAT model provided in an embodiment of the present application. Figure 3 As shown, the method includes the following steps:

[0057] Step 301: Update the first node representation through the graph attention network (GAT) model to obtain the fourth node representation of the current turbine.

[0058] In some embodiments of the present application, the graph attention network GAT model can use a deep learning method to simultaneously integrate the feature information and structural information of the network, and the first node is represented by the following formula: i Represents an update to obtain the fourth node representation h′ of the current turbine i .

[0059] h′ (i,j) =f edge (h i ,h j ,x (i,j) )

[0060]

[0061] Where h′(i,j) represents the updated hidden representation of the edge between the current turbine i and its upstream neighbor j in the neural network, h i is the first node representation of the current turbine i (i.e., the hidden representation of the first node in the neural network), h j is the second node representation of the upstream neighbor j. As the initial node representation, let h i =x i The message passing update is determined by the structure of the graph and is performed to obtain the updated fourth node representation h′ i and edge representation h′ (i,j) , N i is the number of upstream neighbors of the current turbine.

[0062] Step 302 : Obtain an attention coefficient between the current turbine and its upstream neighbor according to the fourth node representation and the second node representation.

[0063] In some embodiments of the present application, the similarity coefficient between the current turbine and its upstream neighbor can be obtained according to the fourth node representation and the second node representation by the following formula:

[0064]

[0065] Among them, e ij is the similarity coefficient between the current turbine i and its upstream neighbor j, a is a single-layer feedforward neural network, W is the weight matrix, is the fourth node, The second node is represented by .

[0066] like Figure 4 As shown in Figure 2, after obtaining the similarity coefficient, the similarity coefficient between the current turbine and its upstream neighbors is further normalized using the softmax function to obtain the attention coefficient between the current turbine and its upstream neighbors. The formula for obtaining the attention coefficient is expressed as follows.

[0067] α ij =softmax j (e ij )

[0068] Among them, α ij is the attention coefficient between the current turbine i and its upstream neighbor j, e ij is the similarity coefficient between the current turbine i and its upstream neighbor j.

[0069] Step 303 : Perform weighted summation based on the attention coefficient between the current turbine and its upstream neighbor, the fourth node representation, and the second node representation to obtain a third node representation of the current turbine.

[0070] In some embodiments of the present application, in order to learn richer information, a multi-head attention mechanism can be adopted. According to the attention coefficient between the current turbine and its upstream neighbors, the fourth node representation and the second node representation, a weighted sum is performed using the following formula to obtain the third node representation of the current turbine.

[0071]

[0072] in, After the p-layer network, the updated third node of the current turbine i is represented, K is the number of attention heads, σ(·) is the activation function, α k is the attention coefficient matrix between the current turbine i and its upstream neighbor j, W is the weight matrix, is the vector representation of the current turbine i at layer p, is the vector representation of the current turbine j at the p-1 layer.

[0073] Through the embodiments of the present application, a pre-trained graph attention network GAT model is used to adopt an attention mechanism to describe the degree of influence of different upstream neighbors in the wind farm on the current turbine, assign different attention weights, and obtain a third node representation that can more accurately describe the current turbine.

[0074] Figure 5 This is a schematic diagram of a wind farm power prediction device based on a graph attention mechanism provided in an embodiment of the present application. Figure 5 As shown, the wind farm power generation prediction device based on the graph attention mechanism includes: a determination module 501, a first acquisition module 502, a second acquisition module 503, a third acquisition module 504 and a fourth acquisition module 505.

[0075] Determination module 501 is configured to determine a current turbine and its upstream neighbors. The upstream neighbors are turbines upstream of the current turbine that meet a preset condition, where the angle between the line connecting the upstream turbine and the current turbine and the wind direction of the wind farm is less than a preset threshold.

[0076] The first acquisition module 502 is configured to determine a pre-constructed wind farm network directed graph and obtain characteristic information of the current turbine based on the network directed graph and its upstream neighbors. The characteristic information of the current turbine includes a characteristic distance between the current turbine and its upstream neighbors, and a characteristic angle between the current turbine and its upstream neighbors.

[0077] The second acquisition module 503 is configured to acquire a first node representation of the current turbine and a second node representation of an upstream neighbor according to the distance feature and the angle feature.

[0078] The third acquisition module 504 is configured to obtain a third node representation of the current turbine based on the first node representation, the second node representation, and a preset graph attention (GAT) network model. The GAT network model has learned to assign different weights to the current turbine and its neighboring turbines.

[0079] The fourth acquisition module 505 is configured to obtain the predicted power of the current turbine based on the third node representation and the preset fully connected neural network MLP model, wherein the MLP model has learned the ability to predict turbine power.

[0080] In some embodiments of the present application, the third acquisition module 504 is specifically configured to update the first node representation using a graph attention network (GAT) model to obtain a fourth node representation of the current turbine. Based on the fourth node representation and the second node representation, obtain an attention coefficient between the current turbine and its upstream neighbor. Finally, perform a weighted sum of the attention coefficient between the current turbine and its upstream neighbor, the fourth node representation, and the second node representation to obtain a third node representation of the current turbine.

[0081] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0082] The wind farm power prediction device based on the graph attention mechanism in the embodiments of this application utilizes the graph attention mechanism and adopts deep learning methods for modeling. This assigns different weights to upstream neighbors that influence the current turbine, identifies the upstream turbine that has the greatest influence on the current turbine, and then obtains the predicted power of the current turbine. This improves prediction accuracy while reducing computational costs. This application can, to a certain extent, address the problem of inaccurate wind farm power prediction caused by the wake effect and reduce computational costs.

[0083] In order to implement the above embodiments, the present application also provides an electronic device. Figure 6 This is a structural block diagram of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device 600 may include a memory 601, a processor 602, and a computer program 603 stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program 603, it executes the wind farm power generation prediction method based on the graph attention mechanism described in any of the above embodiments of the present application.

[0084] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the wind farm power generation prediction method based on the graph attention mechanism described in any of the above embodiments of the present application is implemented.

[0085] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0086] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0087] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0088] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0089] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0090] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A wind farm power prediction method based on graph attention mechanism, characterized in that: The following steps are involved: Determining a current turbine and an upstream neighbor of the current turbine; wherein the upstream neighbor is a turbine among upstream turbines of the current turbine that meets a preset condition, wherein the preset condition is that an angle between a line connecting the upstream turbine and the current turbine and a wind direction of the wind farm is less than a preset threshold; Determining a pre-constructed wind farm network directed graph, and obtaining characteristic information of the current turbine based on the network directed graph and the upstream neighbor; wherein the characteristic information of the current turbine includes a distance characteristic between the current turbine and the upstream neighbor, and an angle characteristic between the current turbine and the upstream neighbor; Obtaining a first node representation of the current turbine and a second node representation of the upstream neighbor according to the distance feature and the angle feature; The first node representation is updated using a preset graph attention network (GAT) model to obtain a fourth node representation of the current turbine. An attention coefficient between the current turbine and the upstream neighbor is obtained based on the fourth node representation and the second node representation. A weighted sum is performed based on the attention coefficient, the fourth node representation, and the second node representation to obtain a third node representation of the current turbine. The graph attention network (GAT) model has learned to assign different weights to the current turbine and its different neighboring turbines. The predicted power of the current turbine is obtained according to the third node representation and a preset fully connected neural network MLP model; wherein the MLP model has learned the ability to predict turbine power.

2. The method according to claim 1, characterized in that The graph attention network (GAT) model updates the first node representation using the following formula to obtain the fourth node representation of the current turbine: in, Indicates the updated current turbine With upstream neighbors The representation of the edge between in the neural network, For the current turbine The first node of For the upstream neighbor The second node indicates that is the fourth node representation of the current turbine.

3. The method according to claim 1, characterized in that Obtaining an attention coefficient between the current turbine and an upstream neighbor of the current turbine according to the fourth node representation and the second node representation includes: obtaining a similarity coefficient between the current turbine and an upstream neighbor of the current turbine based on the fourth node representation and the second node representation; The similarity coefficient between the current turbine and the upstream neighbor of the current turbine is normalized by a softmax function to obtain the attention coefficient between the current turbine and the upstream neighbor of the current turbine.

4. The method according to claim 1, wherein The wind farm network directed graph is pre-constructed in the following way: A network directed graph of the wind farm is constructed based on wind farm conditions and wind farm layout; wherein the nodes in the network directed graph represent turbines in the wind farm, and there is an edge between the node corresponding to each turbine and the node corresponding to the upstream neighbor of each turbine, and the direction of the edge is from the upstream neighbor of each turbine to the corresponding turbine.

5. The method according to claim 1, wherein Obtaining the predicted power of the current turbine according to the third node representation and a preset fully connected neural network MLP model includes: determining an airflow velocity of the wind farm; The predicted power of the current turbine is obtained according to the third node representation, the airflow speed of the wind farm and the preset fully connected neural network MLP model.

6. A wind farm power prediction device based on graph attention mechanism, characterized in that: include: a determination module, configured to determine a current turbine and an upstream neighbor of the current turbine; wherein the upstream neighbor is a turbine among upstream turbines of the current turbine that satisfies a preset condition, wherein the preset condition is that an angle between a line connecting the upstream turbine and the current turbine and a wind direction of the wind farm is less than a preset threshold; a first acquisition module, configured to determine a pre-constructed wind farm network directed graph, and acquire characteristic information of the current turbine based on the network directed graph and the upstream neighbor; wherein the characteristic information of the current turbine includes a distance characteristic between the current turbine and the upstream neighbor, and an angle characteristic between the current turbine and the upstream neighbor; a second acquisition module, configured to acquire a first node representation of the current turbine and a second node representation of the upstream neighbor according to the distance feature and the angle feature; a third acquisition module, configured to update the first node representation using a preset graph attention network (GAT) model to obtain a fourth node representation of the current turbine, obtain an attention coefficient between the current turbine and the upstream neighbor based on the fourth node representation and the second node representation, and obtain a third node representation of the current turbine by performing a weighted summation of the attention coefficient, the fourth node representation, and the second node representation; wherein the graph attention network (GAT) model has learned to assign different weights to the current turbine and its different neighboring turbines; A fourth acquisition module is used to obtain the predicted power of the current turbine based on the third node representation and a preset fully connected neural network MLP model; wherein the MLP model has learned the ability to predict turbine power.

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

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images