Wind power plant data reconstruction method based on graph attention mechanism and manifold auto-encoder

Abstract

The invention belongs to the technical field of data reconstruction, and particularly relates to a wind power plant data reconstruction method based on a graph attention mechanism and a manifold auto-encoder, which comprises the following steps of: firstly, constructing a graph structure of a wind power plant site, representing the site as a node, and enabling the node characteristics to comprise wind speed, wind direction, temperature and generated power; edges between the nodes are defined based on geographical proximity or feature similarity; then, dynamically distributing weights of neighbor nodes through an attention mechanism by using a graph attention encoder, and generating a low-dimensional embedded representation of each site; manifold constraints are introduced into the low-dimensional embedding space, and the local neighborhood relation and the global geometric consistency are ensured; and finally, the low-dimensional embedding is mapped back to a high-dimensional data space through a graph decoder, and reconstruction of missing data is realized. The total loss function provided by the invention combines a data reconstruction error, manifold constraint loss and a regularization item, and model training is completed by optimizing the total loss function.

Description

Technical Field

[0001] The present invention belongs to the technical field of data reconstruction, and in particular relates to a wind farm data reconstruction method based on a graph attention mechanism and a manifold autoencoder. Background Art

[0002] Wind energy is a vital component of renewable energy. The real-time operation of wind farms relies on accurate monitoring of site-level data such as wind speed, direction, temperature, and power generation. However, due to sensor failures, data transmission issues, or other reasons, some sites may experience missing data, impacting wind farm operational efficiency and forecast accuracy. Traditional data imputation methods (such as interpolation and statistical regression models) struggle to fully exploit the spatial correlation between multiple sites and the nonlinear nature of the data.

[0003] In recent years, graph neural networks (GNNs) have gained application in multi-site collaborative forecasting due to their ability to model graph-structured data. Manifold autoencoders, meanwhile, offer advantages in capturing the low-dimensional manifold properties of data. However, existing technologies have failed to effectively combine the advantages of both, leveraging the spatial and manifold properties of multi-site data to dynamically construct weighted relationships between nodes and address the issue of missing data. Summary of the Invention

[0004] In response to the technical problem of missing wind farm site data in the above-mentioned prior art, the present invention provides a wind farm data reconstruction method based on a graph attention mechanism and a manifold autoencoder, which fully utilizes the spatial correlation between sites and the data manifold structure to achieve high-precision filling of missing data.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: The wind farm data reconstruction method based on the graph attention mechanism and manifold autoencoder includes the following steps: S1. Graph construction module: Represent multiple sites of a wind farm as a graph , where: node set Represents a site in a wind farm. The characteristics of each node include wind speed, wind direction, and temperature. The edge set Representing relationships between sites, edge weights are defined based on geographic proximity or feature similarity; S2, Graph Attention Encoder: Utilizes the graph attention mechanism to extract site features and dynamically assigns weights to neighboring nodes to generate a low-dimensional embedding representation of the node; S3, Manifold Constraint Module: imposes local and global manifold constraints on nodes in a low-dimensional embedding space to preserve the geometric characteristics of the data; S4, graph decoder: maps the low-dimensional embedding representation back to the high-dimensional data space and reconstructs node features, including filling in missing values; S5. Loss function optimization: Define the total loss function, including data reconstruction loss, local manifold constraint loss, global manifold constraint loss and regularization loss; S6. Model training and inference: Train the model by optimizing the loss function; use the trained model to reconstruct data with missing values in the wind farm.

[0006] The edge set in the graph construction module in S1 The weight is defined based on both geographic proximity and feature similarity, where feature similarity is determined by calculating the Pearson correlation coefficient of the node feature vectors. By dynamically updating the adjacency matrix, time series data can be modeled to adapt to the time-varying characteristics of the relationship between sites.

[0007] The weight calculation formula in S2 is: in, and Node and nodes The eigenvector of For nodes For Node The attention weight, Representation node The set of neighbor nodes of Represents the weight matrix, which is used to perform linear transformation on node features; Represents the attention weight vector, which is used to calculate the correlation between nodes; Represents the feature splicing operation, which splices two vectors by dimension; Representation node For Node The unnormalized attention score of , indicating the correlation between the two; represents the activation function, Represents the exponential function.

[0008] The graph attention encoder in S2 uses a multi-head attention mechanism to calculate dynamic weights, and the formula is: in, is the number of attention heads, Indicates the The attention weight of the head.

[0009] The formulas for the local and global manifold constraints in S3 are: Local manifold constraints: Global manifold constraint: in, and For nodes and nodes The low-dimensional embedding representation of Represents the actual distance between nodes; the local constraints in the manifold constraint module are jointly defined by the weight of the node neighborhood and the low-dimensional embedding distance, giving priority to preserving the geometric relationship between nodes in the neighborhood.

[0010] The global geometric consistency constraint in the manifold constraint module in S3 is based on the actual distance of high-dimensional data. , calculated by geodesic distance or Euclidean distance.

[0011] The S4 image decoder uses a fully connected layer to map the low-dimensional embedding representation Z back to the high-dimensional feature space, and uses the activation function σ to ensure the nonlinear reconstruction capability of the data.

[0012] The method for reconstructing node features in S4 is: in, To reconstruct high-dimensional data, represents the activation function, is the weight matrix of the decoder, is a low-dimensional embedding representation generated by the graph encoder.

[0013] The total loss function in S5 is: in, , Represents the total loss function, which is used to measure the difference between the model prediction and the real data and the satisfaction degree of the constraints; Representation node The original high-dimensional feature vector of Representation node The reconstructed high-dimensional feature vector of 、 、 is the weight hyperparameter, is the weight coefficient of the local constraint, is the weight coefficient of the global constraint, is the regularization strength; represents the data reconstruction loss, represents the local manifold constraint loss, represents the global manifold constraint loss.

[0014] During the model training process of S6, the model parameters are initialized through self-supervised learning to improve the reconstruction performance on the data set with a high missing rate.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This paper presents a method for reconstructing multi-site wind farm data based on a graph attention mechanism and a manifold autoencoder. The graph attention mechanism adaptively adjusts the weights between sites to accurately capture spatial correlations. Combined with the constraints of the manifold autoencoder, this method ensures that low-dimensional embedding preserves the geometric properties of the data. It also demonstrates strong robustness to noise and extreme data loss. Compared to traditional algorithms, this method significantly improves the accuracy and robustness of wind farm data infill. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0017] The structures, proportions, sizes, etc. illustrated in this specification are intended solely to complement the contents disclosed herein and to facilitate understanding and reading by persons skilled in the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall remain within the scope of the technical contents disclosed herein.

[0018] Figure 1 This is a diagram of the overall model architecture provided by Example 1 of the present invention; Figure 2 A schematic diagram of the wind farm site map structure provided in Example 1 of the present invention; Figure 3 This is an architectural diagram of the graph attention encoder provided in Example 1 of the present invention; Figure 4 Schematic diagram of the low-dimensional embedding manifold constraint provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of this application, not all the embodiments. These descriptions are only to further illustrate the features and advantages of the present invention, rather than to limit the claims of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following embodiments are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0021] Example 1: Reconstructing wind speed data at a single site in a wind farm Background: A wind farm has 50 wind turbine sites, each of which records wind speed data. However, due to some sensor failures, about 20% of the site data are randomly missing. This embodiment uses the method of the present invention to reconstruct the missing data, such as Figure 1 As shown, the following steps are included: Step 1. Data preparation: The data source is historical wind speed data from wind farm sites, with a time span of 30 days and a recording frequency of every 10 minutes. Fill missing data with initial placeholders (such as zero values).

[0022] Step 2: Graph construction: Figure 2 As shown, construct the diagram , where: node set Represents 50 stations, and the feature of each node is the wind speed record of the station. Based on geographic proximity, the distance between each site and its five nearest neighbors is calculated. The smaller the distance, the greater the edge weight. Node features are standardized to have a mean of 0 and a standard deviation of 1.

[0023] Step 3: Model design: Figure 3 As shown in the figure, when designing the graph attention encoder (GAT Encoder), a multi-head attention mechanism is used to calculate the dynamic weights between nodes. The low-dimensional embedding representation of each node is generated by the following formula: in, represents the attention weight, is a learnable parameter, is the activation function. Number of attention heads .

[0024] like Figure 4 As shown, local and global manifold constraints are imposed in the low-dimensional embedding space. The local manifold constraint formula is: The global manifold constraint formula is: When designing the graph decoder, the low-dimensional embedding representation ZZZ is mapped back to the high-dimensional space to complete the reconstruction of the missing data. The decoding formula is: Step 5. Model training: The optimization objective is the total loss function: The parameter settings are as follows: 、 、 , Adam optimizer is used, the learning rate is set to 0.001, and the number of training iterations is 500.

[0025] Step 6. Result Evaluation: Use mean square error (MSE) and mean absolute error (MAE) to evaluate the reconstruction effect. Compared with traditional interpolation methods (such as linear interpolation), the method in this embodiment reduces MSE by 35% and MAE by 28%.

[0026] Example 2: Multi-site collaborative wind speed and power generation data filling Background: 100 sites at a wind farm record wind speed, wind direction, and power generation data. Due to extreme weather, some sites have missing wind speed and power data, resulting in a 25% missing rate. This example utilizes the method of the present invention to perform multi-feature collaborative data filling, including the following steps: Step 1: Data Preparation: The data source is the historical records of wind speed, wind direction, and power generation at 100 stations. The feature vector of each node includes wind speed, wind direction, and power generation.

[0027] Step 2: Graph Construction: Build the Graph , where: node set Represents 100 sites. Edge set It is defined based on both geographical proximity and feature similarity, where feature similarity is determined by calculating the Pearson correlation coefficient of node feature vectors.

[0028] Step 3. Model Design: When designing the graph attention encoder (GAT Encoder), a dynamic weight relationship is constructed and multi-feature collaborative modeling is utilized. When designing the manifold constraint module, wind speed and power generation features are simultaneously embedded in a low-dimensional manifold to preserve the correlation between features and between sites. During the decoding process, the missing wind speed and power generation data are reconstructed based on the collaborative relationship between multiple features.

[0029] Step 4: Model training: Parameter setting and optimization are similar to those in Example 1, with the combined loss weight of wind speed and power generation being increased.

[0030] Step 5. Result evaluation: The MSE and MAE of the reconstructed data are reduced by 40% and 32% respectively compared with the traditional regression method.

[0031] The program portion of the technology can be considered a "product" or "article of manufacture" in the form of readable code and / or related data, and is implemented or implemented through computer-readable media. Tangible, permanent storage media can include any memory or storage used by a computer, processor, or similar device or related module. For example, various semiconductor memories, tape drives, disk drives, or any similar device that can provide storage functions for software.

[0032] All or part of the software may sometimes be communicated over a network, such as the Internet or other communication network. Such communication can load the software from one computer device or processor to another. For example: loading from a server or host computer of a video target detection device to a hardware platform of a computer environment, or other computer environment that implements the system, or a system with similar functions related to providing information required for target detection. Therefore, another medium capable of transmitting software elements can also be used as a physical connection between local devices, such as light waves, radio waves, electromagnetic waves, etc., which are transmitted through cables, optical cables or air. Physical media used to carry carriers, such as cables, wireless connections or optical cables and the like, can also be considered as media that carry software. As used herein, unless limited to tangible "storage" media, other terms referring to computer or machine "readable media" refer to media that participate in the process of executing any instructions by the processor.

[0033] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core concept of the present invention. Those skilled in the art should understand that the modules or steps of the present invention can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0034] The above only describes in detail the preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the purpose of the present invention, and various changes should be included in the scope of protection of the present invention.

Claims

1. A wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder, characterized by: The following steps are involved: S1. Graph construction module: Represent multiple sites of a wind farm as a graph , where: node set Represents a site in a wind farm. The characteristics of each node include wind speed, wind direction, and temperature. The edge set Representing relationships between sites, edge weights are defined based on geographic proximity or feature similarity; S2, Graph Attention Encoder: Utilizes the graph attention mechanism to extract site features and dynamically assigns weights to neighboring nodes to generate a low-dimensional embedding representation of the node; S3, Manifold Constraint Module: imposes local and global manifold constraints on nodes in a low-dimensional embedding space to preserve the geometric characteristics of the data; S4, graph decoder: maps the low-dimensional embedding representation back to the high-dimensional data space and reconstructs node features, including filling in missing values; S5. Loss function optimization: Define the total loss function, including data reconstruction loss, local manifold constraint loss, global manifold constraint loss and regularization loss; S6. Model training and inference: Train the model by optimizing the loss function; use the trained model to reconstruct data with missing values in the wind farm.

2. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The edge set in the graph construction module in S1 The weight is defined based on both geographic proximity and feature similarity, where feature similarity is determined by calculating the Pearson correlation coefficient of the node feature vectors. By dynamically updating the adjacency matrix, time series data can be modeled to adapt to the time-varying characteristics of the relationship between sites.

3. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The weight calculation formula in S2 is: in, and Node and nodes The eigenvector of For nodes For Node The attention weight, Representation node The set of neighbor nodes of Represents the weight matrix, which is used to perform linear transformation on node features; Represents the attention weight vector, which is used to calculate the correlation between nodes; Represents the feature splicing operation, which splices two vectors by dimension; Representation node For Node The unnormalized attention score of , indicating the correlation between the two; represents the activation function, Represents the exponential function.

4. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The graph attention encoder in S2 uses a multi-head attention mechanism to calculate dynamic weights, and the formula is: in, is the number of attention heads, Indicates the The attention weight of the head.

5. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The formulas for the local and global manifold constraints in S3 are: Local manifold constraints: Global manifold constraint: in, and For nodes and nodes The low-dimensional embedding representation of Represents the actual distance between nodes; the local constraints in the manifold constraint module are jointly defined by the weight of the node neighborhood and the low-dimensional embedding distance, giving priority to preserving the geometric relationship between nodes in the neighborhood.

6. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The global geometric consistency constraint in the manifold constraint module in S3 is based on the actual distance of high-dimensional data. , calculated by geodesic distance or Euclidean distance.

7. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The S4 image decoder uses a fully connected layer to map the low-dimensional embedding representation Z back to the high-dimensional feature space, and uses the activation function σ to ensure the nonlinear reconstruction capability of the data.

8. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 7 is characterized in that: The method for reconstructing node features in S4 is: in, To reconstruct high-dimensional data, represents the activation function, is the weight matrix of the decoder, is a low-dimensional embedding representation generated by the graph encoder.

9. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1 is characterized in that: The total loss function in S5 is: in, , Represents the total loss function, which is used to measure the difference between the model prediction and the real data and the satisfaction degree of the constraints; Representation node The original high-dimensional feature vector of Representation node The reconstructed high-dimensional feature vector of 、 、 is the weight hyperparameter, is the weight coefficient of the local constraint, is the weight coefficient of the global constraint, is the regularization strength; represents the data reconstruction loss, represents the local manifold constraint loss, represents the global manifold constraint loss.

10. The wind farm data reconstruction method based on graph attention mechanism and manifold autoencoder according to claim 1, characterized in that: During the model training process of S6, the model parameters are initialized through self-supervised learning to improve the reconstruction performance on the data set with a high missing rate.

Images