Training method of wind power prediction model and wind power prediction method

Abstract

The present application provides a training method for a wind power prediction model and a wind power prediction method, the method comprising: obtaining sample data, wherein the sample data comprises sample meteorological data and sample operation data of a wind farm; inputting the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of a wind power prediction result corresponding to the sample data; obtaining a true value of the wind power prediction result corresponding to the sample data; and iteratively training the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained. Thus, the present application combines the temporal features and spatial features of the data to construct a new multivariate graph neural network model structure for model training, thereby obtaining the temporal and spatial correlation between the features of wind power prediction and improving the accuracy and reliability of the wind power prediction results.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a training method for a wind power prediction model and a wind power prediction method. Background Art

[0002] Deep learning models are widely used in the prediction field. In related technologies, convolutional neural network models, recurrent neural network models or long short-term memory network models are often used for wind power prediction. However, based on the above models, the deep spatiotemporal characteristics of wind farm data are not combined, resulting in low accuracy of wind power prediction results. Therefore, how to improve the accuracy and efficiency of wind power prediction results has become an urgent problem to be solved. Summary of the invention

[0003] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0004] According to the first aspect of the present application, a training method for a wind power prediction model is provided, comprising: obtaining sample data, wherein the sample data includes sample meteorological data and sample operation data of a wind farm; inputting the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of a wind power prediction result corresponding to the sample data; obtaining a true value of the wind power prediction result corresponding to the sample data; and iteratively training the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained.

[0005] According to the second aspect of the present application, a wind power prediction method is provided, comprising: obtaining real-time meteorological data and real-time operation data of a wind farm to be detected; inputting the real-time meteorological data and real-time operation data into a target multivariate graph neural network model to obtain a wind power prediction result of the wind farm to be detected, wherein the target multivariate graph neural network model is a model trained by the model training method described in the first aspect.

[0006] According to a third aspect of the present application, a training device for a wind power prediction model is provided, comprising: a first acquisition module for acquiring sample data, wherein the sample data includes sample meteorological data and sample operation data of a wind farm; a second acquisition module for inputting the sample data into a multivariate graph neural network model to be trained for wind power prediction, so as to obtain a predicted value of a wind power prediction result corresponding to the sample data; a third acquisition module for acquiring a true value of a wind power prediction result corresponding to the sample data; and a training module for iteratively training the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained.

[0007] According to the fourth aspect of the present application, a wind power prediction device is provided, comprising: an acquisition module for acquiring real-time meteorological data and real-time operation data of a wind farm to be detected; a determination module for inputting the real-time meteorological data and real-time operation data into a target multivariate graph neural network model to determine the wind power prediction result of the wind farm to be detected, wherein the target multivariate graph neural network model is a model trained using the model training method described in the first aspect.

[0008] In order to achieve the above-mentioned purpose, the fifth aspect of the present application proposes an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, it implements the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect.

[0009] In order to achieve the above-mentioned objectives, the sixth aspect of the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect.

[0010] In order to achieve the above-mentioned purpose, the seventh aspect of the present application proposes a computer program product, including a computer program, which, when executed by a processor, implements the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect.

[0011] The technical solution provided by the embodiments of the present application includes at least the following beneficial effects:

[0012] The present application provides a training method for a wind power prediction model. By combining the temporal and spatial characteristics of the data and constructing a new multivariate graph neural network model structure for model training, the temporal and spatial correlation between the characteristics of wind power prediction can be obtained, thereby improving the accuracy and reliability of the wind power prediction results.

[0013] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application.

[0015] Figure 1 A schematic diagram of a flow chart of a method for training a wind power prediction model provided in an embodiment of the present application;

[0016] Figure 2 A schematic flow chart of another method for training a wind power prediction model provided in an embodiment of the present application;

[0017] Figure 3 A schematic flow chart of another method for training a wind power prediction model provided in an embodiment of the present application;

[0018] Figure 4 A schematic flow chart of another method for training a wind power prediction model provided in an embodiment of the present application;

[0019] Figure 5 A schematic flow chart of another method for training a wind power prediction model provided in an embodiment of the present application;

[0020] Figure 6 A schematic diagram for comparing a model wind power prediction value and an actual value provided in an embodiment of the present application;

[0021] Figure 7 A schematic diagram of a model loss function value provided in an embodiment of the present application;

[0022] Figure 8 A schematic diagram of a flow chart of a wind power prediction method provided in an embodiment of the present application;

[0023] Fig. 9 A schematic diagram of the structure of a training device for a wind power prediction model provided in an embodiment of the present application;

[0024] Fig.10 A schematic diagram of the structure of a wind power prediction device provided in an embodiment of the present application;

[0025] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0027] The following uses an embodiment to describe in detail the training method of the wind power prediction model of the present application.

[0028] Figure 1 A schematic flow chart of a method for training a wind power prediction model provided in an embodiment of the present application.

[0029] like Figure 1 As shown, the training method of the wind power prediction model proposed in this embodiment specifically includes the following steps:

[0030] S101. Acquire sample data, where the sample data includes sample meteorological data and sample operation data of a wind farm.

[0031] It should be noted that the present application does not limit the specific method for obtaining sample data, which can be selected according to actual conditions.

[0032] Optionally, historical meteorological data and historical operating data can be obtained from the historical database of the wind farm at preset time intervals, and the historical meteorological data and historical operating data can be used as initial sample data. The correlation of the initial sample data with the actual operating power of the wind farm can be used as sample data based on the correlation.

[0033] S102: input the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of a wind power prediction result corresponding to the sample data.

[0034] In an embodiment of the present application, the model structure of the multivariate graph neural network model can be pre-constructed, and the model structure of the multivariate graph neural network model can be composed of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer and a prediction layer.

[0035] In an embodiment of the present application, after the sample data is obtained, the sample data is input into a multivariate graph neural network model to be trained for wind power prediction, and a predicted value of a wind power prediction result corresponding to the sample data can be obtained.

[0036] S103: Obtain the true value of the wind power prediction result corresponding to the sample data.

[0037] In the embodiment of the present application, after the sample data is obtained, the true value of the wind power prediction result corresponding to the sample data can be obtained.

[0038] S104. According to the predicted value of the wind power result and the true value of the wind power prediction result, the multivariate graph neural network model is iteratively trained until a converged target multivariate graph neural network model is obtained.

[0039] In an embodiment of the present application, after obtaining the predicted value of the wind power result and the true value of the wind power prediction result, the loss function of the multivariate graph neural network model can be obtained according to the predicted value of the wind power result and the true value of the wind power prediction result, and the model parameters of the multivariate graph neural network model can be adjusted according to the loss function until the loss function meets the training end conditions, and the multivariate graph neural network model after the last adjustment of the model parameters is determined as the converged target multivariate graph neural network model.

[0040] Optionally, the loss function may be a mean square error (MSE) loss function, a root mean square error (RMSE) to reduce the difference between a predicted value of a wind power prediction result and a true value of the wind power prediction result.

[0041] It should be noted that the present application does not limit the setting of the training end condition, and it can be selected according to the actual situation.

[0042] Optionally, the training end condition can be set as the loss function value being less than a preset loss threshold.

[0043] Optionally, the training end condition can be set as the number of times the parameters of the multivariate graph neural network model are adjusted reaches a preset threshold number of times.

[0044] The training method of the wind power prediction model provided in the present application obtains sample data, wherein the sample data includes sample meteorological data and sample operation data of the wind farm, and inputs the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of the wind power prediction result corresponding to the sample data, obtains a true value of the wind power prediction result corresponding to the sample data, and iteratively trains the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained. The present application combines the temporal characteristics and spatial characteristics of the data to construct a new multivariate graph neural network model structure for model training, which can obtain the temporal and spatial correlation between the characteristics of wind power prediction, thereby improving the accuracy and reliability of the wind power prediction results.

[0045] It should be noted that in the present application, when attempting to obtain sample data, it can be obtained based on historical meteorological data and historical operation data of the wind farm.

[0046] As a possible implementation, Figure 2 As shown, based on the above steps, the specific process of obtaining sample data in the above step S101 includes the following steps:

[0047] S201. Acquire historical meteorological data and historical operation data of the wind farm to form initial sample data based on the historical meteorological data and historical operation data.

[0048] It should be noted that the present application does not limit the specific method for obtaining the historical meteorological data and historical operation data of the wind farm, which can be selected according to actual conditions.

[0049] Optionally, historical meteorological data and historical operation data may be acquired from a historical database of the wind farm at preset time intervals.

[0050] For example, when the preset time length is 15 minutes, historical meteorological data and historical operation data may be obtained from the historical database of the wind farm every 15 minutes to form initial sample data.

[0051] Optionally, the historical meteorological data includes, but is not limited to, data such as wind speed, wind direction, temperature, air pressure and humidity.

[0052] Optionally, the historical operation data includes but is not limited to actual operating power of the wind farm, operating wind speed and other data.

[0053] S202: Obtain the correlation between the initial sample data and the actual operating power of the wind farm.

[0054] It should be noted that the present application does not limit the specific method for obtaining the correlation between the initial sample data and the actual operating power of the wind farm, and it can be selected according to the actual situation.

[0055] Optionally, the correlation between the initial sample data and the actual operating power of the wind farm may be obtained according to a Pearson correlation coefficient calculation formula.

[0056] S203. Select sample data from the initial sample data based on the correlation.

[0057] In the embodiment of the present application, after the correlation is obtained, sample data can be selected from the initial sample data based on the correlation.

[0058] Optionally, a preset correlation threshold may be obtained, and if the correlation between the initial sample data and the actual operating power of the wind farm is greater than the correlation threshold, the initial sample data is determined as the sample data.

[0059] It should be noted that the present application does not limit the setting of the preset correlation threshold, which can be set according to actual conditions.

[0060] Optionally, the initial sample data may be sorted in descending order of relevance, and the first N initial sample data are determined as sample data, where N is a positive integer.

[0061] It should be noted that the present application does not limit the number of N, which can be set according to actual conditions.

[0062] The training method of the wind power prediction model provided in the present application obtains historical meteorological data and historical operating data of the wind farm to form initial sample data based on the historical meteorological data and historical operating data, obtains the correlation between the initial sample data and the actual operating power of the wind farm, and selects sample data from the initial sample data based on the correlation. By obtaining the correlation, the present application can screen the initial sample data, thereby improving the reliability of obtaining the sample data and laying the foundation for accurately obtaining the wind power prediction model.

[0063] It should be noted that after obtaining the sample data, the sample data can be input into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of the wind power prediction result corresponding to the sample data.

[0064] As a possible implementation, Figure 3 As shown, based on the above steps, in the above step S102, the sample data is input into the multivariate graph neural network model to be trained for wind power prediction to obtain the specific process of the wind power prediction result prediction value corresponding to the sample data, including the following steps:

[0065] S301. Input the sample data into a multivariate graph neural network model, wherein the multivariate graph neural network model is composed of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer, and a prediction layer.

[0066] It should be noted that Convolutional Neural Networks (CNN) and Gate Recurrent Unit (GRU) are parallel structures, and temporal graph neural networks and spatial graph neural networks are parallel structures.

[0067] S302. Obtain spatial graph data and temporal graph data of sample data through a convolutional neural network and a gated recurrent unit.

[0068] It should be noted that a graph is a data structure that can describe the relationship between nodes. Graph neural networks have achieved great success in processing graph data due to their permutation invariance, local connectivity, and combinatoriality. From a graph perspective, multivariate time series prediction can be viewed as nodes that are interconnected through hidden layer dependencies. Therefore, graph neural networks can model multivariate time series prediction, construct spatial graph networks and temporal graph networks from the perspective of graph neural networks, and use graph neural networks to solve the problem of multivariate time series prediction.

[0069] As a possible implementation, Figure 4 As shown, based on the above steps, the specific process of obtaining the spatial graph data and the temporal graph data of the sample data through the convolutional neural network and the gated recurrent unit in the above step S302 includes the following steps:

[0070] S401. Extracting spatial features of sample data through a convolutional neural network to obtain spatial features of the sample data, and extracting temporal features of the sample data through a gated recurrent unit to obtain temporal features of the sample data.

[0071] It should be noted that the convolutional neural network is used as a spatial feature extractor, and the structure of the convolutional neural network mainly includes a convolution layer, a rectified linear unit (ReLU) excitation layer, a pooling layer, and a fully connected layer.

[0072] It should be noted that the gated recurrent unit is used as a temporal feature extractor, where the gated recurrent unit layer can capture the dependency of features in the temporal dimension. The gated recurrent unit structure includes setting the number of hidden units to h, the given time step to t, the input to the output of the previous layer, and the hidden state of the previous time step to x t , the calculation formulas of reset gate and update gate are as follows:

[0073] R t =σ(x t w xr +h t-1 w hr +b r )

[0074] z t =σ(x t w xz +h t-1 w hz +b z )

[0075] Among them, R t is the reset gate output, z t is the update gate output, σ is the activation function, w xz 、w xz 、w hz 、w hr is the weight matrix, b r , b z is the offset.

[0076] is a candidate hidden state, and the update mechanism of the hidden state is:

[0077]

[0078]

[0079] Among them, w xh and w hh is the weight matrix, b h is the offset.

[0080] S402: Acquire spatial feature correlation data of spatial features and temporal correlation data of temporal features.

[0081] In the embodiment of the present application, after the spatial features and the temporal features are acquired, the spatial feature correlation data of the spatial features and the temporal correlation data of the temporal features may be acquired.

[0082] Among them, the spatial feature correlation data refers to the correlation coefficient data between two spatial features, and the temporal feature correlation data refers to the correlation coefficient data between two temporal features.

[0083] It should be noted that the present application does not limit the specific method of obtaining spatial feature correlation data and temporal feature correlation data, which can be selected according to actual conditions.

[0084] Optionally, the spatial feature correlation data can be calculated based on the correlation coefficient calculation formula:

[0085]

[0086] Among them, r(X,Y) is the spatial feature correlation data of spatial feature X and spatial feature Y, X and Y are spatial features, cov(X,Y) is covariance, and var[X] and var[Y] represent variance.

[0087] Optionally, the time feature correlation data may be calculated based on the correlation coefficient calculation formula:

[0088]

[0089] Among them, r(A,B) is the time feature correlation data of time feature A and time feature B, A and B are time features, cov(A,B) is covariance, and var[A] and var[B] are variances.

[0090] S403: Use the spatial features and spatial feature correlation data as spatial graph data, and use the temporal features and temporal feature correlation data as temporal graph data.

[0091] It should be noted that after obtaining spatial features and spatial feature correlation data, the spatial features and spatial feature correlation data can be used as spatial graph data, that is, for each sample data, each spatial feature is used as a node of the graph, and the correlation coefficients between each pair of spatial features are used as edges as spatial graph data.

[0092] It should be noted that after obtaining the time features and the time feature correlation data, the time features and the time feature correlation data can be used as time graph data, that is, for each sample data, each time feature is used as a node of the graph, and the correlation coefficients between the two time features are used as edges as time graph data.

[0093] S303. Process the time graph data and the space graph data respectively through the time graph neural network and the space graph neural network to obtain the time graph neural network output and the space graph neural network output.

[0094] It should be noted that the structure of the spatial graph neural network introduces an attention mechanism in the traditional graph neural network to learn the contribution of neighbor nodes to the central node. Different from the graph attention neural network, the calculation of the attention mechanism in this application requires fewer parameters. The formula for calculating the attention mechanism is as follows:

[0095]

[0096] in, is the activation function, p i and p j is the latent space feature of nodes i and j with different spatial features, and score is the scoring function used for pi and p j The dot product calculation of .

[0097] The output h of the spatial graph neural network layer i ’ It can be expressed as:

[0098]

[0099] Among them, σ is the activation function, R ij Calculation result of attention mechanism, Wh j is the weight parameter.

[0100] It should be noted that the constructed time graph data is input into the time graph neural network, and the output of the time graph neural network can be expressed as:

[0101] H l+1 =σ(·)(AH l W l )

[0102] Among them, W l is the weight parameter matrix of the lth layer, σ(·) is the nonlinear activation function (e.g. ReLU), H l+1 is the output of the temporal graph neural network, which is associated with all its neighboring nodes through node features, that is, it can be expressed as the adjacency matrix A and the feature H l Multiply.

[0103] S304. The temporal graph neural network output and the spatial graph neural network output are fused through a fully connected fusion layer to obtain a fully connected fusion layer output.

[0104] It should be noted that the fully connected fusion layer is the time-space fusion layer, and the output of the fully connected fusion layer can be expressed as:

[0105] y=(x1+x2)w xp +b p

[0106] Among them, x1 is the output of the temporal graph neural network, x2 is the output of the spatial graph neural network, y is the output of the fully connected fusion layer, and w xp is the weight, b p is the offset.

[0107] S305. Process the output of the fully connected fusion layer through the spatiotemporal mapping layer and the prediction layer to obtain a predicted value of the wind power prediction result.

[0108] The specific process of the training method of the wind power prediction model of the present application is explained below.

[0109] For example, the model structure of the wind power prediction model (i.e., multivariate graph neural network model) is as follows: Figure 5 As shown in the figure, it consists of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer and a prediction layer. The sample data is obtained by performing correlation analysis on the initial number of samples, and the obtained sample data is input into the convolutional neural network and the gated recurrent unit respectively. The convolutional neural network is used as a spatial feature extractor to capture the spatial correlation between different features, and the representation is constructed based on the spatial graph neural network (i.e., learning spatial correlation). The gated recurrent unit is used as a temporal feature extractor to capture the temporal relationship between different features, and the representation is constructed based on the temporal graph neural network, i.e., learning temporal correlation, and the temporal relationship is extracted. The spatial and time-space are combined, the spatial graph neural network layer is used to learn the spatial graph structure, and the temporal graph neural network layer is used to learn the temporal graph structure to form spatiotemporal association, so as to better extract key information to improve the accuracy of wind power prediction results. The temporal graph neural network output and the spatial graph neural network output are fused through the fully connected fusion layer to obtain the fully connected fusion layer output, and the fully connected fusion layer output is processed through the spatiotemporal mapping layer and the prediction layer to obtain the predicted value of the wind power prediction result. The mean square error is used as the loss function, and the model is iteratively trained to adjust the model parameters until the model converges to obtain the target multivariate graph neural network model.

[0110] Furthermore, the real data and predicted values ​​of wind power can be obtained to determine the predicted effect of wind power. For example, the real data and predicted values ​​of the transformer model are as follows: Figure 6-1 As shown in Figure 2, the actual data and predicted values ​​of the BiLSTM model are as follows: Figure 6-2 As shown in the figure, the real data and predicted values ​​of the Lightweight Gradient Boosting Machine Model (Light GBM) are as follows Figure 6-3 As shown in Figure 2, the actual data and predicted values ​​of the loss function value of the single deep graph convolutional network model (GNN only) are as follows Figure 6-4 As shown, the real data and predicted values ​​of the target multivariate graph neural network model (method of the present invention) are as follows Figure 6-5 shown.

[0111] It should be noted that through Figure 6-1 to Figure 6-5Analysis shows that: the transformer model is not sensitive enough to features and has poor prediction effect. The network structure of the transformer model is more suitable for medium- and long-term prediction. The bidirectional long short-term memory network model cannot learn the oscillation changes of the power curve well, and the bidirectional long short-term memory network is more suitable for single-step output. The prediction effect of the lightweight gradient boosting machine model is poor, and the predicted power is consistent with the approximate range of the true power, but the details cannot match the changes in the true value well. The single deep graph convolutional network model is an idea that only uses GNN. In the later stage, the vibration amplitude of the prediction curve is large and the model is not very stable. The multivariate graph neural network model provided by the present invention performs best in the comparison between the predicted value and the true value. The time graph structure reflects the time information of the data from the perspective of time series, and the spatial graph structure explores the spatial characteristics of the data. The multivariate graph neural network is used to improve the prediction ability of the model.

[0112] Furthermore, a data set of any wind farm can be selected and divided into a training set, a validation set and a test set according to a preset division ratio. During the model training process, a loss function loss graph of the training set train and the validation set val is generated as a function of the number of iterations. The loss function loss graph can reflect the convergence of the model. For example, Figure 7-1 As shown in Figure 2, the single deep convolutional network model only uses the basic graph neural network and has converged when the loss value of the training set train is 0.06, and the model no longer learns, as shown in Figure 2. Figure 7-2 As shown in the figure, the loss values ​​of the training sets train and val of the transformer model drop rapidly in the early stage of training, and then no longer change significantly with iterations in the later stage. The loss value is basically stable at 0.08. It can be seen that the transformer model is not suitable for short-term prediction tasks, such as Figure 7-3 As shown in the figure, the bidirectional long short-term memory network reaches convergence before the number of iterations reaches 100. The model structure is relatively simple, and the loss value is stable at around 0.050. The lightweight gradient boosting machine model generally does not use the loss function graph, which will not be repeated here. Figure 7-4 As shown in the figure, the multivariate graph neural network model basically reaches convergence before the number of iterations reaches 90, and the loss value is the best converged among the selected models, which is basically stable at 0.005. The temporal graph neural network used in the model plays an important role in temporal feature learning, and the spatial graph neural network used in the model plays an important role in the spatial analysis of features. The combination of multivariate graph neural networks can better improve the prediction effect.

[0113] The training method of the wind power prediction model provided in the present application is through inputting sample data into a multivariate graph neural network model, wherein the multivariate graph neural network model is composed of a convolutional neural network, a gated recurrent unit, a time graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer and a prediction layer, and the spatial graph data and the time graph data of the sample data are obtained through the convolutional neural network and the gated recurrent unit, and the time graph data and the spatial graph data are processed respectively through the time graph neural network and the spatial graph neural network to obtain the time graph neural network output and the spatial graph neural network output, and the time graph neural network output and the spatial graph neural network output are fused through the fully connected fusion layer to obtain the fully connected fusion layer output, and the fully connected fusion layer output is processed through the spatiotemporal mapping layer and the prediction layer to obtain the predicted value of the wind power prediction result. The present application can capture the spatiotemporal correlation between different characteristics of wind power prediction through the multivariate graph neural network model, thereby improving the accuracy and reliability of the wind power prediction results.

[0114] It should be noted that after obtaining the trained target multivariate graph neural network model, wind power can be predicted based on the target multivariate graph neural network model.

[0115] The wind power prediction method of the present application is described in detail below using an embodiment.

[0116] Figure 8 A schematic diagram of the flow chart of the wind power prediction method provided in the embodiment of the present application.

[0117] like Figure 8 As shown, the wind power prediction method proposed in this embodiment specifically includes the following steps:

[0118] S801. Acquire real-time meteorological data and real-time operation data of the wind farm to be detected.

[0119] The wind farm to be detected may be any wind farm for which wind power prediction is required.

[0120] Optionally, the historical meteorological data includes, but is not limited to, data such as wind speed, wind direction, temperature, air pressure and humidity.

[0121] Optionally, the historical operation data includes but is not limited to actual operating power of the wind farm, operating wind speed and other data.

[0122] S802: Input the real-time meteorological data and the real-time operation data into the target multivariate graph neural network model to obtain the wind power prediction result of the wind farm to be detected.

[0123] Among them, the target multivariate graph neural network model is the model trained by the above-mentioned model training method.

[0124] In an embodiment of the present application, real-time meteorological data and real-time operation data can be input into a target multivariate graph neural network model for calculation, and a wind power prediction result with a future time length of t can be output.

[0125] The wind power prediction method provided in the present application can obtain real-time meteorological data and real-time operation data of the wind farm to be detected, and input the real-time meteorological data and real-time operation data into the target multivariate graph neural network model to obtain the wind power prediction result of the wind farm to be detected. The present application can input the real-time meteorological data and real-time operation data into the target multivariate graph neural network model, thereby improving the accuracy and reliability of obtaining the wind power prediction results.

[0126] In order to implement the above embodiment, this embodiment provides a training device for a wind power prediction model. Fig. 9 A schematic diagram of the structure of a training device for a wind power prediction model provided in an embodiment of the present application.

[0127] like Fig. 9 As shown, the training device 1000 of the wind power prediction model includes: a first acquisition module 110 , a second acquisition module 120 , a third acquisition module 130 and a training module 140 .

[0128] A first acquisition module 110, configured to acquire sample data, wherein the sample data includes sample meteorological data and sample operation data of the wind farm;

[0129] A second acquisition module 120 is used to input the sample data into a multivariate graph neural network model to be trained for wind power prediction, so as to obtain a predicted value of a wind power prediction result corresponding to the sample data;

[0130] A third acquisition module 130 is used to obtain a true value of the wind power prediction result corresponding to the sample data;

[0131] The training module 140 is used to iteratively train the multivariate graph neural network model according to the wind power prediction result value and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained.

[0132] According to one embodiment of the present application, the first acquisition module 110 is also used to: obtain historical meteorological data and historical operating data of the wind farm to form initial sample data based on the historical meteorological data and the historical operating data; obtain the correlation between the initial sample data and the actual operating power of the wind farm; and select the sample data from the initial sample data based on the correlation.

[0133] According to one embodiment of the present application, the second acquisition module 120 is also used to: input the sample data into the multivariate graph neural network model, wherein the multivariate graph neural network model is composed of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer and a prediction layer; obtain the spatial graph data and the temporal graph data of the sample data through the convolutional neural network and the gated recurrent unit; process the temporal graph data and the spatial graph data through the temporal graph neural network and the spatial graph neural network, respectively, to obtain the temporal graph neural network output and the spatial graph neural network output; fuse the temporal graph neural network output and the spatial graph neural network output through the fully connected fusion layer to obtain the fully connected fusion layer output; process the fully connected fusion layer output through the spatiotemporal mapping layer and the prediction layer to obtain the predicted value of the wind power prediction result.

[0134] In one embodiment of the present application, the second acquisition module 120 is further used to: extract the spatial features of the sample data through the convolutional neural network to obtain the spatial features of the sample data, and extract the temporal features of the sample data through the gated recurrent unit to obtain the temporal features of the sample data; obtain spatial feature correlation data of the spatial features and temporal correlation data of the temporal features; use the spatial features and the spatial feature correlation data as the spatial graph data, and use the temporal features and the temporal feature correlation data as the temporal graph data.

[0135] In one embodiment of the present application, the training module 140 is further used to: obtain the loss function of the multivariate graph neural network model based on the wind power prediction value and the true value of the wind power prediction result; adjust the model parameters of the multivariate graph neural network model according to the loss function until the loss function meets the training end condition, and determine the multivariate graph neural network model after the last adjustment of the model parameters as the converged target multivariate graph neural network model.

[0136] The training device of the wind power prediction model provided in the present application obtains sample data, wherein the sample data includes sample meteorological data and sample operation data of the wind farm, and inputs the sample data into the multivariate graph neural network model to be trained for wind power prediction to obtain the predicted value of the wind power prediction result corresponding to the sample data, obtains the true value of the wind power prediction result corresponding to the sample data, and iteratively trains the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained. Thus, by combining the temporal characteristics and spatial characteristics of the data and constructing a new multivariate graph neural network model structure for model training, the temporal and spatial correlation between the characteristics of the wind power prediction can be obtained, thereby improving the accuracy and reliability of the wind power prediction results.

[0137] In order to implement the above embodiment, this embodiment provides a wind power prediction device. Fig.10 A schematic diagram of the structure of a wind power prediction device provided in an embodiment of the present application.

[0138] like Fig.10 As shown, the wind power prediction device 2000 includes: an acquisition module 210 and a determination module 220 .

[0139] An acquisition module 210 is used to acquire real-time meteorological data and real-time operation data of the wind farm to be detected;

[0140] The determination module 220 is used to input the real-time meteorological data and the real-time operation data into the target multivariate graph neural network model to determine the wind power prediction result of the wind farm to be detected, wherein the target multivariate graph neural network model is a model trained using the model training method described in the first aspect.

[0141] The wind power prediction device provided in the present application can obtain real-time meteorological data and real-time operation data of the wind farm to be detected, and input the real-time meteorological data and real-time operation data into the target multivariate graph neural network model to obtain the wind power prediction result of the wind farm to be detected. The present application can input real-time meteorological data and real-time operation data into the target multivariate graph neural network model, thereby improving the accuracy and reliability of obtaining wind power prediction results.

[0142] In order to implement the above embodiment, the present application also proposes an electronic device 3000, such as Fig.11 As shown, it includes: a memory 310, a processor 320, and a computer program stored in the memory 310 and executable on the processor 320. When the processor executes the program, the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect is implemented.

[0143] In order to implement the above embodiments, the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect.

[0144] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which, when executed by a processor, implements the training method of the wind power prediction model described in the first aspect or the wind power prediction method described in the second aspect.

[0145] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0146] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A training method for a wind power prediction model, characterized in that: The method comprises: Acquiring sample data, wherein the sample data includes sample meteorological data and sample operation data of the wind farm; Inputting the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of a wind power prediction result corresponding to the sample data; Obtaining a true value of a wind power prediction result corresponding to the sample data; Iteratively training the multivariate graph neural network model according to the wind power prediction result value and the wind power prediction result true value until a converged target multivariate graph neural network model is obtained; The step of inputting the sample data into a multivariate graph neural network model to be trained for wind power prediction to obtain a predicted value of a wind power prediction result corresponding to the sample data includes: Input the sample data into the multivariate graph neural network model, wherein the multivariate graph neural network model consists of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer, and a prediction layer, the convolutional neural network and the gated recurrent unit are parallel structures, and the temporal graph neural network and the spatial graph neural network are parallel structures; Obtaining spatial graph data and temporal graph data of the sample data through a convolutional neural network and a gated recurrent unit; The time graph data and the space graph data are processed by the time graph neural network and the space graph neural network respectively to obtain a time graph neural network output and a space graph neural network output; The temporal graph neural network output and the spatial graph neural network output are fused through the fully connected fusion layer to obtain a fully connected fusion layer output; Processing the output of the fully connected fusion layer through the spatiotemporal mapping layer and the prediction layer to obtain a predicted value of the wind power prediction result; The obtaining of the spatial graph data and the temporal graph data of the sample data by using a convolutional neural network and a gated recurrent unit includes: Extracting the spatial features of the sample data by the convolutional neural network to obtain the spatial features of the sample data, and extracting the temporal features of the sample data by the gated recurrent unit to obtain the temporal features of the sample data; Acquire spatial feature correlation data of the spatial feature and temporal correlation data of the temporal feature; The spatial features and the spatial feature correlation data are used as the spatial graph data, and the temporal features and the temporal feature correlation data are used as the temporal graph data.

2. The method according to claim 1, characterized in that The obtaining of sample data comprises: Acquiring historical meteorological data and historical operation data of the wind farm to form initial sample data according to the historical meteorological data and the historical operation data; Obtaining the correlation between the initial sample data and the actual operating power of the wind farm; The sample data is selected from the initial sample data according to the correlation.

3. The method according to claim 1, characterized in that The iterative training of the multivariate graph neural network model according to the wind power prediction result value and the wind power prediction result true value until a converged target multivariate graph neural network model is obtained includes: Obtaining a loss function of the multivariate graph neural network model according to the predicted value of the wind power result and the true value of the wind power prediction result; The model parameters of the multivariate graph neural network model are adjusted according to the loss function until the loss function meets the training end condition, and the multivariate graph neural network model after the last adjustment of the model parameters is determined as the converged target multivariate graph neural network model.

4. A wind power prediction method, comprising: Obtain real-time meteorological data and real-time operation data of the wind farm to be tested; The real-time meteorological data and real-time operation data are input into a target multivariate graph neural network model to obtain a wind power prediction result of the wind farm to be detected, wherein the target multivariate graph neural network model is a model trained using the model training method described in any one of claims 1 to 3.

5. A training device for a wind power prediction model, characterized in that: The device comprises: A first acquisition module, used to acquire sample data, wherein the sample data includes sample meteorological data and sample operation data of the wind farm; A second acquisition module is used to input the sample data into a multivariate graph neural network model to be trained for wind power prediction, so as to obtain a predicted value of a wind power prediction result corresponding to the sample data; A third acquisition module is used to obtain the true value of the wind power prediction result corresponding to the sample data; A training module, used for iteratively training the multivariate graph neural network model according to the wind power prediction result value and the true value of the wind power prediction result until a converged target multivariate graph neural network model is obtained; The second acquisition module is further used to input the sample data into the multivariate graph neural network model, wherein the multivariate graph neural network model is composed of a convolutional neural network, a gated recurrent unit, a temporal graph neural network, a spatial graph neural network, a fully connected fusion layer, a spatiotemporal mapping layer and a prediction layer, the convolutional neural network and the gated recurrent unit are parallel structures, and the temporal graph neural network and the spatial graph neural network are parallel structures; Obtaining spatial graph data and temporal graph data of the sample data through a convolutional neural network and a gated recurrent unit; The time graph data and the space graph data are processed by the time graph neural network and the space graph neural network respectively to obtain a time graph neural network output and a space graph neural network output; The temporal graph neural network output and the spatial graph neural network output are fused through the fully connected fusion layer to obtain a fully connected fusion layer output; Processing the output of the fully connected fusion layer through the spatiotemporal mapping layer and the prediction layer to obtain a predicted value of the wind power prediction result; The obtaining of the spatial graph data and the temporal graph data of the sample data by using a convolutional neural network and a gated recurrent unit includes: Extracting the spatial features of the sample data by the convolutional neural network to obtain the spatial features of the sample data, and extracting the temporal features of the sample data by the gated recurrent unit to obtain the temporal features of the sample data; Acquire spatial feature correlation data of the spatial feature and temporal correlation data of the temporal feature; The spatial features and the spatial feature correlation data are used as the spatial graph data, and the temporal features and the temporal feature correlation data are used as the temporal graph data.

6. A wind power prediction device, characterized in that: The device comprises: An acquisition module is used to acquire real-time meteorological data and real-time operation data of the wind farm to be detected; A determination module is used to input the real-time meteorological data and real-time operation data into a target multivariate graph neural network model to determine the wind power prediction result of the wind farm to be detected, wherein the target multivariate graph neural network model is a model trained using the model training method described in any one of claims 1 to 3.

7. An electronic device, characterized in that: including a processor and a memory; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the method according to any one of claims 1 to 3 or claim 4.

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 3 or claim 4 is implemented.

Images