Short-term wind power output prediction method and related device

Abstract

The invention belongs to a prediction method, and provides a short-term wind power output prediction method and a related device for solving the technical problems that the short-term and ultra-short-term prediction effect is poor, the prediction precision is reduced along with time extension and the prediction result precision is low in the current wind power output prediction. The historical wind power data set is input into a short-term wind power output prediction model, a wind power prediction result is obtained, and the short-term wind power output prediction model comprises an adjacency matrix module used for conducting similarity calculation on all features in the wind power data set to obtain an initialized adjacency matrix, and a spatial feature extraction module used for extracting spatial features according to the initialized adjacency matrix; the spatial feature extraction module is used for carrying out spatial feature extraction by adopting graph learning and graph convolution to obtain spatial features, the time feature extraction module is used for inputting the spatial features into a Transform encoder to carry out time feature extraction to obtain a multi-feature prediction result, and the full connection module is used for processing the multi-feature prediction result to obtain a wind power prediction result.

Description

Technical Field

[0001] The present application relates to a prediction method, and specifically to a short-term wind power output prediction method and related devices. Background Art

[0002] Wind power output forecasting is part of the basic work in power grid dispatching and energy distribution. Accurately predicting the output power of wind turbines is crucial to maintaining the safe and stable operation of the power grid and ensuring the balance of power supply and demand.

[0003] At present, wind power output forecasting mainly uses physical modeling methods, statistical modeling methods and artificial intelligence algorithms. Among them, the physical modeling method relies on a lot of domain expertise and is usually suitable for long-term forecasting, but is not effective in short-term and ultra-short-term forecasting; the statistical modeling method builds a statistical model by analyzing historical data, which is suitable for processing linear relationships and does not consider meteorological information such as wind speed and wind direction. The prediction accuracy decreases with time; the artificial intelligence algorithm can capture nonlinear relationships and has high prediction accuracy, but the existing model fails to conduct comprehensive predictions in the field of wind power combined with time and space dimensions, resulting in low prediction accuracy and instability in practical applications. Summary of the invention

[0004] The present application aims to provide a short-term wind power output prediction method and related devices to address the technical problems that the current wind power output prediction has poor results in short-term and ultra-short-term predictions, the prediction accuracy decreases with time, and the prediction result accuracy is low.

[0005] In order to achieve the above objectives, this application adopts the following technical solutions: In the first aspect, the present application proposes a short-term wind power output prediction method, comprising: Get historical wind power datasets; Inputting the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result; The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initialization adjacency matrix; The spatial feature extraction module is used to extract spatial features by using graph learning and graph convolution according to the initialization adjacency matrix to obtain spatial features; A temporal feature extraction module, used to input the spatial features into a Transformer encoder for temporal feature extraction to obtain a multi-feature prediction result; The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results.

[0006] Furthermore, after obtaining the historical wind power data set, the method further includes: performing normalization processing on the historical wind power data set.

[0007] Furthermore, the normalization of the historical wind power data set includes: The historical wind power data set is resampled at different time granularities, the average data value within the historical time period is calculated, the processed data is aggregated, the missing values ​​are backfilled, and the data is normalized.

[0008] Furthermore, the similarity calculation of each feature in the wind power data set includes:

[0009] in, Features and Features The similarity calculation results are: Features The value at all timestamps, Features The value at all timestamps, is the bandwidth parameter.

[0010] Furthermore, the method of extracting spatial features by using graph learning and graph convolution based on the initialized adjacency matrix includes: Through a learnable parameter matrix, the initial adjacency matrix is ​​adjusted according to the characteristics of the wind power data set, the weights between the nodes in the initial adjacency matrix are learned, the adjacency matrix is ​​generated, and the corresponding degree matrix is ​​calculated; Based on the degree matrix, the adjacency matrix is ​​symmetrically normalized to obtain the normalized result; The wind power dataset and normalized results are passed to the graph convolutional network, and the spatial features are captured through a multi-layer graph convolutional network.

[0011] Furthermore, the Transformer encoder includes six encoding layers connected in sequence, each encoding layer including a multi-head self-attention mechanism and a feedforward neural network; Before the spatial features are input into the Transformer encoder for temporal feature extraction, the method further includes: concatenating the spatial features with the timestamp to generate an input embedding, and adding a position code to the input embedding based on a sine function and a cosine function to obtain an encoded input embedding.

[0012] Furthermore, the operations performed in each coding layer include: For each attention head, the encoded input embedding is linearly transformed to generate the query, key, and value; Calculate the similarity between the query and the key through the dot product to get the attention score matrix; The attention scores in the attention score matrix are converted into attention weights through the Softmax nonlinear activation function; Integrate the attention weights obtained from all attention heads and obtain the multi-head attention output matrix through a linear transformation weight matrix; Combine the multi-head attention output matrix with the wind power data set, add residual connection, and then perform layer normalization to obtain the first layer normalization result; The first layer normalization processing result is input into the feedforward network, the output of the feedforward network is combined with the first layer normalization processing result, a residual connection is added, and the layer normalization processing is performed again to obtain the second layer normalization processing result as the output result of the current coding layer.

[0013] In the second aspect, the present application proposes a short-term wind power output prediction system, comprising: Data module, used to obtain historical wind power data sets; A prediction module, used for inputting the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result; The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initialization adjacency matrix; The spatial feature extraction module is used to extract spatial features by using graph learning and graph convolution according to the initialization adjacency matrix to obtain spatial features; A temporal feature extraction module, used to input the spatial features into a Transformer encoder for temporal feature extraction to obtain a multi-feature prediction result; The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results.

[0014] In a third aspect, the present application proposes an electronic device, comprising: a memory, and one or more processors; the memory is coupled to the processor; wherein computer program code is stored in the memory, and the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the steps of the above-mentioned short-term wind power output forecasting method.

[0015] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned short-term wind power output prediction method are implemented.

[0016] Compared with the prior art, this application has the following beneficial effects: The present application proposes a short-term wind power output prediction method, constructs a short-term wind power output prediction model, and can predict wind power based on historical wind power data sets, wherein the short-term wind power output prediction model includes an adjacency matrix module, a spatial feature extraction module, a temporal feature extraction module, and a fully connected module, and the historical wind power data set can include a variety of feature data, comprehensively considering the various factors affecting wind power output, so that the model can capture more complex nonlinear relationships, and significantly improve the accuracy of the prediction compared to the method of using only a single feature. The spatiotemporal dimension analysis method with multiple time granularities is adopted, so that the model can capture the changes in wind power data from different time scales. This flexibility allows the model to adapt to short-term and long-term change trends, and is more robust and adaptable than traditional methods. Combined with the above-mentioned multi-factor considerations and multi-time granularity analysis, the present application can greatly improve the deviation and uncertainty problems existing in the existing prediction model, improve the stability and accuracy of the prediction results, and is of great significance for optimizing wind power scheduling, reducing wind abandonment, and promoting the effective use of renewable energy.

[0017] The present application also proposes a short-term wind power output prediction system, an electronic device and a computer storage medium, which have all the advantages of the above-mentioned short-term wind power output prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a schematic diagram of the first flow chart of the short-term wind power output prediction method of this application; Figure 2 This is a second flow chart of the short-term wind power output prediction method of this application; Figure 3 This is a schematic diagram of the overall architecture of the short-term wind power output prediction model; Figure 4 This is a comparison chart of the prediction results at a time granularity of 30 minutes in the embodiment of the present application; Figure 5 A schematic diagram of the short-term wind power output prediction system of this application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0022] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0023] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0024] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0025] In the description of the embodiments of the present application, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0026] Wind power output forecasting is the core link of power grid dispatching and energy management, and is of great significance for ensuring stable operation of the power grid and the balance of power supply and demand. However, the wind power of wind farms is complex and changeable, showing nonlinear and non-stationary characteristics. At the same time, the forecasting model is also deeply affected by time and space factors. In order to effectively integrate wind power and other meteorological information and realize the power generation forecast of wind farms, it is necessary to identify the inertia, non-stationarity, periodicity and multi-scale characteristics of wind, and select appropriate models based on different characteristics.

[0027] At present, wind power output prediction technology mainly revolves around three major directions: physical modeling, statistical modeling, and artificial intelligence algorithms. Among them, the physical modeling method, with its deep domain knowledge support, can show excellent performance in long-term prediction. This method builds a sophisticated prediction model by deeply understanding the physical characteristics of the wind farm, such as terrain, wind turbine layout, aerodynamics, etc. However, when facing short-term and ultra-short-term predictions, the physical modeling method often seems to be powerless, because its complex calculation process and reliance on a large amount of professional knowledge limit its ability to respond quickly to wind changes. The statistical modeling method focuses on mining patterns from historical data and predicting future wind power output by building statistical models. This method performs well in dealing with linear relationships and can quickly give prediction results. However, a significant drawback of the statistical modeling method is that it often ignores the impact of key meteorological information such as wind speed and wind direction on wind power output, resulting in a gradual decrease in prediction accuracy over time. Especially when the wind changes drastically or the meteorological conditions are complex, the predictive ability of the statistical model may be greatly reduced. In contrast, artificial intelligence algorithms have shown great potential in wind power output forecasting, being able to capture complex nonlinear relationships, process massive amounts of data, and have the ability to self-learn and optimize. This enables artificial intelligence algorithms to often surpass traditional physical modeling and statistical modeling methods in terms of forecasting accuracy. However, existing artificial intelligence models are still not mature enough for specific needs in the wind power sector, and often fail to fully combine the time and space dimensions for all-round forecasting, resulting in the stability and accuracy of the forecast results still needing to be further improved.

[0028] In order to give full play to the advantages of artificial intelligence algorithms in wind power output forecasting, future research needs to explore the characteristics of wind power data more deeply, develop more intelligent prediction models, and continuously optimize the performance of the algorithm to meet the power industry's demand for accurate predictions.

[0029] Based on the above situation, the present application proposes a short-term wind power output prediction method and related devices, and the present application is described in detail below in conjunction with embodiments and drawings.

[0030] like Figure 1 As shown, it is a first flow chart of the short-term wind power output prediction method of the present application, which may include: S101, obtaining a historical wind power data set.

[0031] Collecting and organizing wind power data from wind farms over a period of time can serve as the basis for subsequent predictions. In practical applications, historical wind power data sets can include various meteorological factors such as the actual power generation of the wind farm, wind speed, wind direction, temperature, humidity, and possible timestamp information. When training the short-term wind power output prediction model in the future, historical wind power data sets are also required. These data can cover a sufficiently long period of time to ensure that the model can learn long-term characteristics such as the periodicity and seasonality of wind power output. In addition, before entering the short-term wind power output prediction model, the data can also be preprocessed, including cleaning outliers, filling missing values, normalization or standardization, etc., to ensure the quality and consistency of the data.

[0032] S102: Input the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result.

[0033] The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initial adjacency matrix. By constructing the similarity relationship between the features in the wind power data set, it provides a basis for the subsequent spatial feature extraction. In practical applications, appropriate similarity measurement methods (such as Euclidean distance, cosine similarity, etc.) can be used to calculate the similarity between the features in the wind power data set (such as wind speed and wind direction at different time points). Then, the initial adjacency matrix is ​​constructed according to the similarity value, where the elements in the initial adjacency matrix represent the connection strength or similarity between the features.

[0034] The spatial feature extraction module is used to extract spatial features based on the initialized adjacency matrix using graph learning and graph convolution to obtain spatial features. It should be noted that the spatial correlation between wind turbines or measuring points within the wind farm can be captured by extracting spatial features from the initialized adjacency matrix using graph learning and graph convolution techniques. Through graph learning algorithms (such as graph neural networks) and graph convolution operations, the spatial relationship of wind power data is modeled and features are extracted, which can capture the spatial correlation between wind turbines or measuring points at different locations in the wind farm due to factors such as terrain and meteorological conditions, providing rich spatial feature information for subsequent predictions.

[0035] The temporal feature extraction module is used to input the spatial features into the Transformer encoder for temporal feature extraction to obtain multi-feature prediction results. This application utilizes the powerful sequence modeling capabilities of the Transformer encoder to extract the temporal features of wind power data and capture the time series laws of wind power output. The spatial feature sequence output by the spatial feature extraction module can be used as the input of the Transformer encoder. Through its unique self-attention mechanism, the Transformer encoder can efficiently process long-term dependencies in sequence data, thereby extracting temporal features in wind power data, such as trends, periodic changes, etc.

[0036] The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results. It should be noted that the fully connected layer linearly combines and nonlinearly activates the input features through the weight matrix and bias term to achieve feature fusion and conversion. The spatial features and temporal features are fused and further processed through the fully connected layer to finally obtain the wind power prediction result, which can be the wind power output value or the change trend of wind power in the future.

[0037] This application uses an adaptive graph convolutional network for spatial feature extraction and a Transformer encoder for temporal feature extraction, which can capture the nonlinear and complex characteristics of wind power data, comprehensively consider multiple factors such as wind power and meteorology, and comprehensively evaluate the impact of environmental conditions on wind power generation.

[0038] like Figure 2 As shown, it is a second flow chart of the short-term wind power output prediction method of the present application, which may include: S201, wind power data set construction and preprocessing.

[0039] This example is based on the training, verification and testing of a wind farm data in northern China. The data set contains 10 features (wind speed and direction at different locations, humidity, air pressure and actual power generation), a total of 35040 sampling points, from 00:00:00 on January 1, 2019 to 23:45:00 on December 31, 2019, with an original time granularity of 15 minutes. Some data are shown in Table 1.

[0040] Table 1 Partial dataset of a wind farm in northern China

[0041] Then, the wind power data set is preprocessed. Specifically, in the preprocessing stage, the original data is resampled at different time granularities, and the average value of the data in the original time period is calculated to aggregate the processed data. Through resampling, the original data is aggregated according to different time granularities (such as hourly, daily, weekly, etc.) to obtain a more macro data trend. And the missing values ​​are backfilled to avoid errors in subsequent analysis or modeling. The data is then normalized to scale the data to a specific range (such as between 0 and 1) to eliminate the influence of different dimensions and make each feature at the same order of magnitude, which is conducive to the convergence of subsequent models.

[0042] S202, constructing a short-term wind power output prediction model.

[0043] In this embodiment, the short-term wind power output prediction model adopts the AGFormer deep learning model. Figure 3 As shown in the figure, it is a schematic diagram of the overall architecture of the short-term wind power output prediction model. The architecture of AGFormer mainly includes a spatial feature extraction module and a temporal feature extraction module. The spatial feature extraction module can effectively capture the mutual influence between nodes through adaptive graph convolution. Adaptive graph convolution includes graph learning and graph convolution processes. During the graph learning process, the structure of the graph will be automatically updated according to the changes in the data. The graph convolution performs spatial aggregation on the features according to the similarity and correlation of different time series, captures global information and local features, and extracts spatial features. The data after spatial feature extraction is passed into the Transformer encoder for temporal feature extraction to obtain multi-feature prediction results; finally, the fully connected layer is used to process the predicted output of wind power.

[0044] Specifically, the following methods may be included: (1) For the preprocessed input sequence First, use the Gaussian kernel function to calculate the similarity based on each variable to get the initial adjacency matrix :

[0045] Treat each feature as one-dimensional time series data. For each pair of features and Features , which is the value of these two features at all timestamps and , apply the Gaussian kernel function to calculate the similarity between variable pairs, is the bandwidth parameter, and the Euclidean distance between all variables can be calculated. , take the median of the distance as For one-dimensional time series data, features are organized in columns and time is organized in rows.

[0046] Then through a learnable parameter matrix , the initial adjacency matrix is ​​adjusted based on 10 characteristic data such as wind speed, wind direction, temperature, air pressure, humidity and power generation , learn the weights between nodes and generate the adjacency matrix , and calculate the corresponding degree matrix .

[0047] (2) Based on the degree matrix, the adjacency matrix is ​​symmetrically normalized to obtain , so that the characteristics of each node can be To effectively propagate. and the adjacency matrix Passed to the graph convolutional network, it captures the deeper spatial features of the sequence through multiple layers of graph convolution. The spatial feature extraction result obtained by combining the hidden state of each layer of graph convolution As the input of Transformer.

[0048] (3) Transformer contains 6 encoding layers, each of which contains two parts: a multi-head self-attention mechanism and a feedforward neural network. First, the spatial features are With timestamp Concatenate to generate input embedding , and adds positional encoding to the input embedding based on sine and cosine functions to preserve sequence information.

[0049] (4) For each attention head, the input embedding Generate query after linear transformation ,key Sum , calculate the similarity between the query and the key through the dot product, and get the attention score matrix through Softmax Non-linear activation function converts attention scores into attention weights . Integrate the outputs of all attention heads through a linear transformation weight matrix Get the multi-head attention output matrix .

[0050] (5) Outputting multi-head attention With input Combined with adding residual connections and performing layer normalization. The normalized results are passed through the feedforward network, and residual connections and normalization are added again to obtain the temporal feature extraction results of each layer. . Six encoder layers are stacked together, and the output of each layer is used as the input of the next layer. The last layer of Transformer encoder outputs the multivariate prediction results .

[0051] (6) Output the multivariate prediction combining temporal and spatial features Through the fully connected layer, the prediction result of wind power output is obtained .

[0052] Afterwards, the wind power forecast was also experimentally analyzed, and the wind power forecast results were analyzed: 1. Wind power prediction experimental analysis.

[0053] Based on the short-term wind power output forecasting model, the wind power data sets at different time granularities are analyzed at prediction step sizes of 12, 24, 48, and 96. The input step size is fixed at 96, and MSE (mean square error) and MAE (mean absolute error) are used as evaluation indicators. The experimental results are shown in Table 2.

[0054] Table 2 Experimental results of wind power prediction at different time granularities of AGFormer

[0055] According to the results in Table 2, the short-term wind power output forecasting model shows excellent forecasting performance at different time granularities. By calculating the average MSE and MAE at each time granularity, it is found that when the granularity is set to 30 minutes, the forecast error is the lowest compared to other time granularities. The model can capture the optimal cycle of multiple variables affecting wind power output and achieve more accurate forecasting. When the time granularity is 30 minutes, the average MSE and MAE are 0.147 and 0.244 respectively. The experimental results further show that for wind power data of different time granularities, combined with multivariate influencing factors, the data can be analyzed and processed through adaptive spatiotemporal modeling, which can significantly improve the accuracy of the forecast. By comparing and analyzing the forecast results at different time granularities, the optimal cycle of wind speed, wind direction and meteorological changes can be found, and the accuracy and stability of wind power output forecasting can be improved.

[0056] 2. Analysis of wind power forecast results.

[0057] Through experimental analysis of a wind power data set in northern China, the most appropriate time granularity is selected: 30 minutes, and the predicted value at any time point in the inference process with a prediction length of 96 hours is visualized. Figure 4 As shown in the figure, it is a comparison chart of the prediction results at a time granularity of 30 minutes. Figure 4 In the figure, the black line represents the true value and the red line represents the predicted value. Figure 3 It can be seen that at the time granularity of 30 minutes, the short-term wind power output prediction model can best capture the wind speed, wind direction and meteorological change cycle, and achieve more accurate wind power generation prediction.

[0058] This application first pre-processes the input wind power data according to the time granularity of the prediction target and the characteristics of wind power changes, and then inputs the pre-processed data into the spatial feature extraction module, constructs a topological map of wind speed, wind force, meteorology and other features based on the adaptive graph convolutional network, analyzes the correlation between variables and extracts spatial features. The results of the spatial feature extraction module are then input into the temporal feature extraction module, and the temporal dependency of each variable in different subspaces is extracted based on the multi-head attention mechanism of the Transformer encoder to obtain a multi-dimensional prediction output. Finally, the required wind power output prediction results are generated through full connection layer processing. Through adaptive spatiotemporal modeling, this application can effectively capture the complex dynamics of wind power data changing with time and space, and improve the accuracy and stability of the prediction.

[0059] like Figure 5 As shown, it is a schematic diagram of a short-term wind power output prediction system of the present application, which may include: Data module, used to obtain historical wind power data sets; A prediction module, used for inputting the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result; The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initialization adjacency matrix; The spatial feature extraction module is used to extract spatial features by using graph learning and graph convolution according to the initialization adjacency matrix to obtain spatial features; A temporal feature extraction module, used to input the spatial features into a Transformer encoder for temporal feature extraction to obtain a multi-feature prediction result; The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results.

[0060] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of each module is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The module described as a separate component may or may not be physically separated. The component displayed as a module may be a physical unit or multiple physical units, that is, it may be located in one place, or it may be distributed in multiple different places. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0061] In addition, each module in each embodiment of the present invention may be integrated into a processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0062] An embodiment of the present application also provides an electronic device, which may include one or more processors, a memory, and a communication interface.

[0063] The memory, the communication interface and the processor are coupled, for example, the memory, the communication interface and the processor may be coupled together via a bus.

[0064] The communication interface is used for data transmission with other devices. The memory stores computer program code. The computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the steps of the above short-term wind power processing prediction method.

[0065] Among them, the processor can be a processor or a controller, for example, a central processing unit (CPU), a general processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the present disclosure. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of DSP and microprocessors, and the like. The processor can be used to support electronic devices to execute the method steps provided in the above embodiments.

[0066] The bus may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The above bus may be divided into an address bus, a data bus, a control bus, etc.

[0067] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned short-term wind power processing prediction method are implemented.

[0068] The computer-readable storage medium involved in the present application includes random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field.

[0069] For the description of the relevant parts of the short-term wind power processing prediction system, electronic device and computer-readable storage medium provided in the embodiments of the present application, please refer to the detailed description of the corresponding parts in the short-term wind power processing prediction method provided in the embodiments of the present application, which will not be repeated here. In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0070] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A short-term wind power output prediction method, characterized in that: include: Get historical wind power datasets; Inputting the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result; The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initialization adjacency matrix; The spatial feature extraction module is used to extract spatial features by using graph learning and graph convolution according to the initialization adjacency matrix to obtain spatial features; A temporal feature extraction module, used to input the spatial features into a Transformer encoder for temporal feature extraction to obtain a multi-feature prediction result; The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results.

2. The short-term wind power processing prediction method according to claim 1 is characterized in that: After the historical wind power data set is obtained, the method further includes: performing normalization processing on the historical wind power data set.

3. The short-term wind power processing prediction method according to claim 2 is characterized in that: The normalization process of the historical wind power data set includes: The historical wind power data set is resampled at different time granularities, the average data value within the historical time period is calculated, the processed data is aggregated, the missing values ​​are backfilled, and the data is normalized.

4. The short-term wind power processing prediction method according to claim 1, characterized in that: The similarity calculation of each feature in the wind power data set includes: in, Features and Features The similarity calculation results are: Features The value at all timestamps, Features The value at all timestamps, is the bandwidth parameter.

5. The short-term wind power processing prediction method according to claim 4 is characterized in that: The method of extracting spatial features by using graph learning and graph convolution according to the initialized adjacency matrix includes: Through a learnable parameter matrix, the initial adjacency matrix is ​​adjusted according to the characteristics of the wind power data set, the weights between the nodes in the initial adjacency matrix are learned, the adjacency matrix is ​​generated, and the corresponding degree matrix is ​​calculated; Based on the degree matrix, the adjacency matrix is ​​symmetrically normalized to obtain the normalized result; The wind power dataset and normalized results are passed to the graph convolutional network, and the spatial features are captured through a multi-layer graph convolutional network.

6. The short-term wind power processing prediction method according to claim 4, characterized in that: The Transformer encoder includes six encoding layers connected in sequence, each encoding layer includes a multi-head self-attention mechanism and a feedforward neural network; Before the spatial features are input into the Transformer encoder for temporal feature extraction, the method further includes: concatenating the spatial features with the timestamp to generate an input embedding, and adding a position code to the input embedding based on a sine function and a cosine function to obtain an encoded input embedding.

7. The short-term wind power processing prediction method according to claim 6, characterized in that: The operations performed within each encoding layer include: For each attention head, the encoded input embedding is linearly transformed to generate the query, key, and value; Calculate the similarity between the query and the key through the dot product to get the attention score matrix; The attention scores in the attention score matrix are converted into attention weights through the Softmax nonlinear activation function; Integrate the attention weights obtained from all attention heads and obtain the multi-head attention output matrix through a linear transformation weight matrix; Combine the multi-head attention output matrix with the wind power data set, add residual connection, and then perform layer normalization to obtain the first layer normalization result; The first layer normalization processing result is input into the feedforward network, the output of the feedforward network is combined with the first layer normalization processing result, a residual connection is added, and the layer normalization processing is performed again to obtain the second layer normalization processing result as the output result of the current coding layer.

8. A short-term wind power output prediction system, characterized in that: include: Data module, used to obtain historical wind power data sets; A prediction module, used for inputting the historical wind power data set into a short-term wind power output prediction model to obtain a wind power prediction result; The short-term wind power output prediction model includes: The adjacency matrix module is used to calculate the similarity of each feature in the wind power data set to obtain the initialization adjacency matrix; The spatial feature extraction module is used to extract spatial features by using graph learning and graph convolution according to the initialization adjacency matrix to obtain spatial features; A temporal feature extraction module, used to input the spatial features into a Transformer encoder for temporal feature extraction to obtain a multi-feature prediction result; The fully connected module is used to process the multi-feature prediction results to obtain the wind power prediction results.

9. An electronic device, characterized in that: include: A memory and one or more processors; the memory is coupled to the processor; wherein the memory stores computer program code, the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the steps of the short-term wind power output prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the short-term wind power output prediction method according to any one of claims 1 to 7 are implemented.

Images