Image-based wind power short-term prediction method and system with self-attention mechanism and gating

Abstract

The application discloses a kind of with self-attention mechanism and gate image-based wind power short-term prediction method and system, belong to wind power generation prediction technical field, wherein, the method includes: obtaining NWP meteorological data and SCADA fan data is preprocessed, and the data after preprocessing is divided into training set and test set;Training set is decomposed into continuous multiple historical time subsequences, and is reconstructed into two-dimensional image;Residual-based deep convolutional neural network is established, and gate convolutional neural network layer and self-attention mechanism are added, to obtain residual-based deep convolutional neural network with self-attention mechanism and gate;Two-dimensional image is input into the neural network training;Test set is input into the neural network trained and is predicted, to obtain short-term wind power.The method comprehensively uses historical time series data and meteorological grid data, extracts features and converts into images, uses advanced image processing technology to predict wind, greatly improves the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and in particular to a method and system for short-term wind power prediction with a self-attention mechanism and gating based on images. Background Technology

[0002] With rapid economic development and continuously improving living standards, human demand for energy is constantly increasing. However, the use of traditional fossil fuels, such as coal, oil, and natural gas, releases pollution, damages the environment, and contributes to global warming. Furthermore, due to the non-renewable nature and limited reserves of fossil fuels, over-exploitation will lead to the depletion of energy resources. Therefore, renewable energy is needed to solve the energy crisis and environmental problems. Wind energy, a pollution-free and widely distributed renewable energy source, has received widespread global attention.

[0003] Unlike the stable energy supply from fossil fuels, wind power systems typically exhibit non-stationary and nonlinear uncertainties. Meteorological parameters such as temperature, humidity, air pressure, wind direction, and wind speed influence the wind power generated by wind turbines. These factors pose significant challenges to the management and operation of power systems, including wind energy. Research on this issue has shown that accurate wind power forecasting improves the reliability and economic feasibility of these systems by reducing their integration and operating costs. However, accurate forecasting is a difficult process due to the high uncertainty of wind speed and direction. The uncertainty of wind data complicates the model learning process, leading to large prediction errors. Therefore, wind energy forecasting is considered a challenging task.

[0004] As is well known, wind speed data exhibits non-stationary and non-linear properties. Due to the cubic relationship between wind speed and wind force, the uncertainty level of power data is higher than that of wind speed data. Therefore, wind power data is decomposed to extract deep feature information from subsequences. Virtual Mode Decomposition (VMD) is a non-recursive decomposition algorithm with a sound theoretical foundation that can extract modes simultaneously. In their paper "An improved residual-based convolutional neural network for very short-term wind power forecasting," Ceyhun Yildiz et al. extracted features from time series using VMD and then reconstructed the features of historical time series into RGB color space images. Transforming time series into visual patterns offers an advantage: the inherently spatially invariant 2D-CNN visual representation provides optimal input for convolutional layers. Using images as input and output, residual-based deep learning models can be established, such as... Figure 1 As shown, the model has low complexity and low computational cost, can be based on multiple learnable parameters, and can produce more effective prediction results.

[0005] However, this model only uses one type of data, and it extracts and converts eight hours of input data into one image. Moreover, the selection of images when constructing the training and test sets is random, which means that the model ignores the very important seasonality of wind speed and wind force and does not take into account the temporal relationship between images. Therefore, although the deep learning model's final prediction is effective, it has a large error. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems in the related art.

[0007] Therefore, the first objective of this invention is to propose a short-term wind power prediction method with self-attention mechanism and gating based on images, which greatly improves the prediction accuracy.

[0008] The second objective of this invention is to propose a short-term wind power prediction system with a self-attention mechanism and gating based on images.

[0009] The third objective of this invention is to provide a short-term wind power prediction device.

[0010] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0011] To achieve the above objectives, a first aspect of the present invention proposes a short-term wind power prediction method based on an image with self-attention and gating, comprising the following steps: Step S1, acquiring NWP meteorological data and SCADA wind turbine data, preprocessing them, and dividing the preprocessed data into a training set and a test set; Step S2, decomposing the training set into multiple consecutive historical time subsequences, and reconstructing the multiple historical time subsequences into a two-dimensional image; Step S3, establishing a residual-based deep convolutional neural network, and adding a gating convolutional neural network layer and a self-attention mechanism to obtain a residual-based deep convolutional neural network with self-attention and gating; Step S4, inputting the two-dimensional image into the residual-based deep convolutional neural network with self-attention and gating for training; Step S5, inputting the test set into the trained residual-based deep convolutional neural network with self-attention and gating for prediction to obtain short-term wind power.

[0012] The wind power short-term prediction method based on images with self-attention mechanism and gating in this invention has the following steps: First, feature extraction is based on variational mode decomposition, and these features are converted into images. Second, a residual-based deep convolutional neural network with self-attention mechanism and gating is used to predict wind power. This model uses meteorological wind speed, wind direction, and wind force data as the dataset, which comes from a combination of numerical weather prediction (NWP) and wind turbine data collected by SCADA. That is, it comprehensively uses historical time series data and meteorological grid data, extracts features and converts them into images, and uses advanced image processing technology to predict wind force, which greatly improves the prediction accuracy.

[0013] In addition, the short-term wind power prediction method with self-attention mechanism and gating based on images according to the above embodiments of the present invention may also have the following additional technical features:

[0014] Furthermore, in one embodiment of the present invention, the SCADA fan data includes wind power, 10-minute average wind speed, rotor speed and fan status, and the NWP meteorological data includes 100-meter horizontal and vertical wind, temperature and humidity within a 7*7 grid centered on the wind farm's location coordinates.

[0015] Furthermore, in one embodiment of the present invention, the preprocessing procedure of step S1 is as follows:

[0016] Interpolation is used to refine the time granularity of the NWP meteorological data, aligning the NWP meteorological data with the SCADA wind turbine data in time steps, and then feature stitching is performed.

[0017] Further, in one embodiment of the present invention, step S2 specifically includes: step S201, using variational mode decomposition (VMD) to decompose the training set into the plurality of historical time subsequences; step S202, using a minimum-maximum normalization method to normalize the wind power data in the plurality of historical time subsequences to ensure that all features are distributed at the same scale, and updating the plurality of historical time subsequences; step S203, reconstructing the updated plurality of historical time subsequences into the two-dimensional image, wherein the two-dimensional image includes hue, saturation, and value maps, and each two-dimensional image is composed of a 2×4 array with a depth of 8 bits.

[0018] To achieve the above objectives, a second aspect of the present invention proposes a short-term wind power prediction system based on an image with self-attention and gating, comprising: an acquisition and preprocessing module for acquiring NWP meteorological data and SCADA wind turbine data, performing preprocessing, and dividing the preprocessed data into a training set and a test set; a reconstruction module for decomposing the training set into multiple consecutive historical time subsequences and reconstructing the multiple historical time subsequences into a two-dimensional image; a construction module for establishing a residual-based deep convolutional neural network and adding a gating convolutional neural network layer and a self-attention mechanism to obtain a residual-based deep convolutional neural network with self-attention and gating; a training module for inputting the two-dimensional image into the residual-based deep convolutional neural network with self-attention and gating for training; and a prediction module for inputting the test set into the trained residual-based deep convolutional neural network with self-attention and gating for prediction to obtain short-term wind power.

[0019] The wind power short-term prediction system based on an image with self-attention and gating according to embodiments of the present invention has the following stages: First, feature extraction is based on variational mode decomposition, and these features are converted into images. Second, a residual-based deep convolutional neural network with self-attention and gating is used to predict wind power. This model uses meteorological wind speed, wind direction, and wind force data as the dataset, which comes from a combination of numerical weather prediction (NWP) and wind turbine data collected based on SCADA. That is, it comprehensively uses historical time-series data and meteorological grid data, extracts features and converts them into images, and uses advanced image processing technology to predict wind force, which greatly improves the prediction accuracy.

[0020] In addition, the wind power short-term prediction system with self-attention mechanism and gating based on image according to the above embodiments of the present invention may also have the following additional technical features:

[0021] Furthermore, in one embodiment of the present invention, the SCADA fan data includes wind power, 10-minute average wind speed, rotor speed and fan status, and the NWP meteorological data includes 100-meter horizontal and vertical wind, temperature and humidity within a 7*7 grid centered on the wind farm's location coordinates.

[0022] Furthermore, in one embodiment of the present invention, the preprocessing process of the acquisition and preprocessing module is as follows: the NWP meteorological data is refined in terms of time granularity using an interpolation method, so that the NWP meteorological data and the SCADA wind turbine data are aligned in time step, and feature stitching is performed.

[0023] Furthermore, in one embodiment of the present invention, the reconstruction module is specifically used to: decompose the training set into the plurality of historical time subsequences using variational mode decomposition (VMD); normalize the wind power data in the plurality of historical time subsequences using a minimum-maximum normalization method to ensure that all features are distributed at the same scale, and update the plurality of historical time subsequences; reconstruct the updated plurality of historical time subsequences into the two-dimensional image, wherein the two-dimensional image includes hue, saturation, and value maps, and each two-dimensional image is composed of a 2×4 array with a depth of 8 bits.

[0024] To achieve the above objectives, a third aspect of the present invention provides a short-term wind power prediction device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the short-term wind power prediction method with self-attention mechanism and gating based on images as described in the above embodiments.

[0025] To achieve the above objectives, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the short-term wind power prediction method with self-attention mechanism and gating based on images as described in the above embodiments.

[0026] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0028] Figure 1 The diagram below is a schematic representation of the structure of a prior art residual-based deep learning model according to an embodiment of the present invention.

[0029] Figure 2 This is a flowchart of an embodiment of the present invention of a short-term wind power prediction method with self-attention mechanism and gating based on image;

[0030] Figure 3 This is a schematic diagram of the structure of a residual-based deep convolutional neural network according to an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram illustrating the execution of the entire gated convolutional neural network layer according to an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram illustrating the execution of a self-attention mechanism according to an embodiment of the present invention;

[0033] Figure 6 This is a flowchart illustrating the prediction of short-term wind power according to an embodiment of the present invention;

[0034] Figure 7 This is a schematic diagram of the structure of a short-term wind power prediction system with self-attention mechanism and gating based on image according to an embodiment of the present invention. Detailed Implementation

[0035] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0036] The following describes, with reference to the accompanying drawings, a method and system for short-term wind power prediction based on an image with self-attention mechanism and gating according to an embodiment of the present invention. First, the method for short-term wind power prediction based on an image with self-attention mechanism and gating according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0037] Figure 2 This is a flowchart of an embodiment of the present invention for a short-term wind power prediction method with self-attention mechanism and gating based on images.

[0038] like Figure 2 As shown, the short-term wind power prediction method with self-attention mechanism and gating based on images includes the following steps:

[0039] In step S1, NWP meteorological data and SCADA wind turbine data are acquired and preprocessed, and the preprocessed data is divided into training set and test set.

[0040] Furthermore, in one embodiment of the present invention, the SCADA fan data includes wind power, 10-minute average wind speed, rotor speed and fan status, and the NWP meteorological data includes 100-meter horizontal and vertical wind, temperature and humidity within a 7*7 grid centered on the location coordinates of the fan site.

[0041] Specifically, the data collected primarily from the manually SCADA system and the NWP numerical meteorological data provided by EC were used. The SCADA data mainly included fields such as wind power, 10-minute average wind speed, rotor speed, and turbine status. EC data used a 7x7 grid centered on the wind farm's coordinates, with a grid size of 0.125 degrees. The main fields used were 100-meter horizontal and vertical wind speeds, temperature, and humidity. For the SCADA data, the 0.5-second resolution data was first divided into 10-minute time spans. The data within each 10-minute span was averaged to obtain the SCADA data for each full 10-minute interval. The averaging method, rather than selecting specific time points, was used to avoid the influence of wind fluctuations on the data; averaging better reflects the wind speed conditions within that time period. Missing time periods were not filled in to prevent data noise from affecting the prediction results. For the NWP data, the NWP data downloaded from EC was in nc format and needed to be read using Python library functions provided on the EC website. After reading the data by coordinates and categories, it was reassembled into a time-sorted tabular data. Finally, the NWP data is transformed into a 3D mesh data with 7*7*feature number.

[0042] Furthermore, in this embodiment of the invention, the preprocessing process is as follows: the NWP meteorological data is refined in terms of time granularity using interpolation, so that the NWP meteorological data and SCADA wind turbine data are aligned in time step, and feature splicing is performed.

[0043] In step S2, the training set is decomposed into multiple consecutive historical time subsequences, and the multiple historical time subsequences are reconstructed into a two-dimensional image.

[0044] Furthermore, in one embodiment of the present invention, step S2 specifically includes:

[0045] Step S201: Variational Mode Decomposition (VMD) is used to decompose the training set into multiple historical time subsequences;

[0046] Step S202: The minimum-maximum normalization method is used to normalize the wind power data in multiple historical time series to ensure that all features are distributed at the same scale and to update multiple historical time series.

[0047] Step S203: The updated multiple historical time subsequences are reconstructed into a two-dimensional image, wherein the two-dimensional image includes hue, saturation and value maps, and each two-dimensional image is composed of a 2×4 array with a depth of 8 bits.

[0048] Specifically, Variational Mode Decomposition (VMD) is used to decompose the time series p(t) of the training set into a series of K eigenmode functions, i.e., multiple historical time subsequences, as shown in the following equation:

[0049]

[0050] Each intrinsic mode function is represented as u k (t), whose amplitude and frequency have the following characteristics:

[0051]

[0052] It should be noted that the VMD algorithm used in this invention is a constrained variational optimization algorithm, which can be expressed by the following formula:

[0053]

[0054]

[0055] Among them, w k Let δ(t) be the center frequency, and δ(t) be the pulse signal. is u k Hilbert transform of (t);

[0056] In other words, the data in the VMD training set is divided into small batches, and the data in each batch must be data from a continuous period of time.

[0057] The obtained historical time sequence is then reconstructed into a two-dimensional image, facilitating learning by a deep neural network in subsequent steps. Reconstructing the features of the historical time sequence into a visual pattern offers the advantage of providing optimal input to 2D CNN visual representation convolutional layers with inherent spatial invariance.

[0058] It should be noted that the two-dimensional images contain hourly input correlations from the power plant. Before the reconstruction process, the wind power data was normalized using a min-max normalization method to ensure that all features are distributed at the same scale. The image format was chosen from the last 8 hours, with each image group consisting of a 2×4 array containing 8 data points. These will be used as input parameters later; the reconstructed features can be defined as f. i 1 ,f i 2 ,L,f i N , where N is the total number of data points, and i is the number of input features, i = 1, 2, ..., 8.

[0059] For each historical time series, such as the decomposed IMF components, the data points are rearranged into an RGB image with a depth of 8 bits, including hue, saturation, and value maps. Each image is generated based on a different historical time series for each input parameter.

[0060] In step S3, a residual-based deep convolutional neural network is established, and a gated convolutional neural network layer and a self-attention mechanism are added to obtain a residual-based deep convolutional neural network with self-attention mechanism and gate.

[0061] Specifically, first, a deep convolutional neural network based on residuals is established, such as... Figure 3 As shown, nonlinear wind power prediction requires extensive deep feature extraction at any wind power frequency. Therefore, this invention constructs a robust and efficient deep residual CNN model that fully utilizes image recognition to extract high-level features from convolutional layers. To achieve efficient aggregation of feature maps, residual connections are used to carry the extracted salient features from the initial layer to deeper layers. The remaining connections also include skip connections to provide additional feature information for more accurate predictions.

[0062] Furthermore, a gated convolutional neural network layer for extracting seasonal features is added to the deep convolutional neural network to learn the wind force variation characteristics during different days. It is important to pay attention to the model input; images should be input sequentially in chronological order. The training set must correspond to the test set: for example, if the wind force needs to be predicted on a specific day in March, the training set should also include wind force data from previous years' March and the months before and after March.

[0063] Therefore, this embodiment of the invention uses a gated convolutional neural network layer to learn the seasonality of wind force, such as Figure 4 As shown, it consists of a one-dimensional convolutional layer and gated linear units. This special design allows for parallel and controllable training. In this process, let the convolutional kernel be Γ, the input be Y, and the output be Z. The output is divided into two equal parts, which can be represented as Z = [P, Q], where [P, Q] serves as the input to the gate. The computation of the entire gated convolutional neural network layer can be represented as:

[0064] Γ*Y=Peσ(Q)

[0065] Where σ(Q) represents the composition and dynamic changes of the learning input P over time.

[0066] Furthermore, such as Figure 5 As shown, by adding a self-attention mechanism, for the same input X... t By generating corresponding Query, Key, and Value through different linear layers, this method allows any point in time within a long sequence to establish sensitive relationships with other points in time, avoiding dependency failures caused by excessively long time windows. This improves the dependency capability of long-term input windows, thus enhancing the model's robustness and generalization ability. In practical applications, it is more easily adapted to the ever-changing wind power generation prediction problem. The above calculation process is expressed by the following formula:

[0067] head = Attention(QW)Q ,KW K VW V ).

[0068] In step S4, the two-dimensional image is input into a residual-based deep convolutional neural network with self-attention mechanism and gating for training.

[0069] In step S5, the test set is input into a trained residual-based deep convolutional neural network with self-attention mechanism and gating to make predictions and obtain short-term wind power.

[0070] It should be noted that the test set data is fed into the network to evaluate the network training results. The data in the test set is complete and does not need to be segmented.

[0071] Specifically, such as Figure 6 As shown, the test set data is input into a trained deep convolutional neural network based on residuals with self-attention and gating. First, the data is input into the deep convolutional neural network based on residuals, and then the result of the deep convolutional neural network based on residuals is fed into the gating neural network layer, allowing the model to learn the wind force variation characteristics (seasonality) during different days. Then, the result of the gating neural network layer is fed into the self-attention layer, and the result of the self-attention layer is added together with the result of the gating neural network layer with weights. The calculation results of the two are then used for regression prediction to obtain the final wind force prediction result, i.e., short-term wind power, and the wind power prediction is performed 4 hours after the reporting point.

[0072] Subsequently, the system obtains real-time SCADA data from the wind turbine manufacturer and real-time NWP data from the EC every day. These data are then fed into the model for prediction, generating prediction data four hours after the reporting point. At the same time, the system feeds the previous day's data into the model for incremental training to enhance the model's performance.

[0073] In summary, the short-term wind power prediction method with self-attention mechanism and gating based on images proposed in the embodiments of the present invention has the following beneficial effects:

[0074] By combining NWP data and SCADA data to decompose and reconstruct 2D images, and by integrating the features of historical time series to reconstruct a 2D CNN visual representation with inherent spatial invariance, the optimal input for the convolutional layers is provided.

[0075] By using gated convolutional neural network layers to learn the relationship between wind force time-series images of different dates, i.e. the seasonality of wind force, the predictive ability of the model is improved.

[0076] The initial training results of the 2D image reconstructed from wind turbine data are fed into the self-attention mechanism. By combining the prediction results of the two parts, the dependency failure problem caused by the excessively long time window is avoided, and the robustness and generalization ability of the model are improved.

[0077] Next, with reference to the accompanying drawings, a short-term wind power prediction system with self-attention mechanism and gating based on images proposed according to an embodiment of the present invention is described.

[0078] Figure 7 This is a flowchart of an embodiment of the present invention of a short-term wind power prediction system with self-attention mechanism and gating based on image.

[0079] like Figure 7 As shown, the system 10 includes: an acquisition and preprocessing module 100, a reconstruction module 200, a construction module 300, a training module 400, and a prediction module 500.

[0080] The system comprises the following modules: Acquisition and Preprocessing Module 100 acquires NWP meteorological data and SCADA wind turbine data, performs preprocessing, and divides the preprocessed data into training and testing sets. Reconstruction Module 200 decomposes the training set into multiple consecutive historical time subsequences and reconstructs these subsequences into a two-dimensional image. Construction Module 300 establishes a residual-based deep convolutional neural network (DCNN) and adds gated DCNN layers and a self-attention mechanism to obtain a residual-based DCNN with self-attention and gated layers. Training Module 400 trains the residual-based DCNN with self-attention and gated layers using the two-dimensional image input. Prediction Module 500 predicts short-term wind power using the test set input into the trained residual-based DCNN with self-attention and gated layers.

[0081] Furthermore, in one embodiment of the present invention, the SCADA fan data includes wind power, 10-minute average wind speed, rotor speed and fan status, and the NWP meteorological data includes 100-meter horizontal and vertical wind, temperature and humidity within a 7*7 grid centered on the location coordinates of the fan site.

[0082] Furthermore, in one embodiment of the present invention, the preprocessing process of the acquisition and preprocessing module is as follows: the NWP meteorological data is refined in terms of time granularity using an interpolation method, so that the NWP meteorological data and SCADA wind turbine data are aligned in time step, and feature splicing is performed.

[0083] Furthermore, in one embodiment of the present invention, the reconstruction module is specifically used to: decompose the training set into multiple historical time subsequences using variational mode decomposition (VMD); normalize the wind power data in the multiple historical time subsequences using a minimum-maximum normalization method to ensure that all features are distributed at the same scale, and update the multiple historical time subsequences; reconstruct the updated multiple historical time subsequences into a two-dimensional image, wherein the two-dimensional image includes hue, saturation, and value maps, and each two-dimensional image is composed of a 2×4 array with a depth of 8 bits.

[0084] It should be noted that the foregoing explanation of the embodiment of the wind power short-term prediction method with self-attention mechanism and gating based on image also applies to the system of this embodiment, and will not be repeated here.

[0085] The short-term wind power prediction system with self-attention mechanism and gating based on image proposed in the embodiments of the present invention has the following beneficial effects:

[0086] By combining NWP data and SCADA data to decompose and reconstruct 2D images, and by integrating the features of historical time series to reconstruct a 2D CNN visual representation with inherent spatial invariance, the optimal input for the convolutional layers is provided.

[0087] By using gated convolutional neural network layers to learn the relationship between wind force time-series images of different dates, i.e. the seasonality of wind force, the predictive ability of the model is improved.

[0088] The initial training results of the 2D image reconstructed from wind turbine data are fed into the self-attention mechanism. By combining the prediction results of the two parts, the dependency failure problem caused by the excessively long time window is avoided, and the robustness and generalization ability of the model are improved.

[0089] To implement the above embodiments, the present invention also proposes a short-term wind power prediction device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the short-term wind power prediction method with self-attention mechanism and gating based on images as described in the foregoing embodiments.

[0090] To implement the above embodiments, the present invention also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the short-term wind power prediction method with self-attention mechanism and gating based on images as described in the foregoing embodiments.

[0091] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0093] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0094] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0095] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0096] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0097] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0098] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for image-based wind power short-term prediction with self-attention mechanism and gating, characterized in that, The method comprises the following steps: Step S1, preprocessing NWP meteorological data and SCADA wind turbine data, and dividing the preprocessed data into a training set and a test set; Step S2, decomposing the training set into a plurality of continuous historical time subsequences, and reconstructing the plurality of historical time subsequences into a two-dimensional image; Step S3, establishing a residual-based deep convolutional neural network, and adding a gated convolutional neural network layer and a self-attention mechanism to obtain a residual-based deep convolutional neural network with a self-attention mechanism and a gate; Step S4, inputting the two-dimensional image into the residual-based deep convolutional neural network with a self-attention mechanism and a gate for training; Step S5, inputting the test set into the trained residual-based deep convolutional neural network with a self-attention mechanism and a gate for prediction to obtain short-term wind power; The SCADA wind turbine data includes wind power, 10-minute average wind speed, wind wheel speed and wind turbine state, and the NWP meteorological data includes 100-meter horizontal and vertical wind, temperature and humidity in a 7*7 grid centered on the coordinates of the wind farm; The step S2 specifically comprises: Step S201, decomposing the training set into the plurality of historical time subsequences by using a variational mode decomposition VMD; Step S202, normalizing wind power data in the plurality of historical time subsequences by using a min-max normalization method to ensure that all features are distributed according to the same scale, and updating the plurality of historical time subsequences; Step S203, reconstructing the updated plurality of historical time subsequences into the two-dimensional image, wherein the two-dimensional image includes a hue, a saturation and a value map, and each two-dimensional image is composed of a 2*4 array in turn, with a depth of 8 bits.

2. The method of claim 1, wherein the method is a wind power short-term prediction method with self-attention mechanism and gated image-based. The preprocessing process of step S1 is: The NWP meteorological data is refined in time granularity by using an interpolation method, so that the NWP meteorological data is aligned with the SCADA wind turbine data in time steps, and feature splicing is performed.

3. An image-based wind power short-term prediction system with self-attention mechanism and gating, characterized in that, It comprises: An acquisition and preprocessing module for acquiring and preprocessing NWP meteorological data and SCADA wind turbine data, and dividing the preprocessed data into a training set and a test set; A reconstruction module for decomposing the training set into a plurality of continuous historical time subsequences, and reconstructing the plurality of historical time subsequences into a two-dimensional image; A construction module for establishing a residual-based deep convolutional neural network, and adding a gated convolutional neural network layer and a self-attention mechanism to obtain a residual-based deep convolutional neural network with a self-attention mechanism and a gate; A training module for inputting the two-dimensional image into the residual-based deep convolutional neural network with a self-attention mechanism and a gate for training; A prediction module for inputting the test set into the trained residual-based deep convolutional neural network with a self-attention mechanism and a gate for prediction to obtain short-term wind power; The SCADA wind turbine data includes wind power, 10-minute average wind speed, wind wheel speed and wind turbine state, and the NWP weather data includes 100-meter horizontal and vertical wind, temperature and humidity within a 7*7 grid centered on the coordinates of the wind farm; The reconstruction module is specifically configured to: The training set is decomposed into the plurality of historical time subsequences by using a variational mode decomposition (VMD); The wind power data in the plurality of historical time subsequences is normalized by using a min-max normalization method to ensure that all features are distributed according to the same scale, and the plurality of historical time subsequences are updated; The updated plurality of historical time subsequences are reconstructed into the two-dimensional images, wherein the two-dimensional images include hue, saturation and value maps, and each two-dimensional image is composed of a 2*4 array in turn, and the depth is 8 bits.

4. The system with self-attention mechanism and gated image-based wind power short-term prediction system according to claim 3, characterized in that, The preprocessing process of the acquisition and preprocessing module is as follows: The NWP weather data is refined in time granularity by using an interpolation method, so that the NWP weather data is aligned with the SCADA wind turbine data in time steps, and feature splicing is performed.

5. A wind power short-term prediction device, characterized by, The computer program is stored in the memory and executable on the processor, and when the processor executes the computer program, the image-based short-term wind power prediction method with self-attention mechanism and gating is realized.

6. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is stored in the memory and executable on the processor, and when the processor executes the computer program, the image-based short-term wind power prediction method with self-attention mechanism and gating is realized.

Images