Wind power prediction model training method, wind power prediction method, device and medium

Abstract

The present invention provides a wind power prediction model training method, wind power prediction method, device, and medium. The training method includes: obtaining time-series wind turbine data of a wind turbine and meteorological numerical data of the wind farm where the wind turbine is located, wherein the meteorological numerical data represents the meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time; performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data; fusing the first feature data and the second feature data to obtain training data; and using the training data to train a pre-built deep neural network to obtain a wind power prediction model. The technical solution of the present invention improves the accuracy of wind power prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a wind power prediction model training method, a wind power prediction method, a device and a medium. Background Art

[0002] With the annual increase in electricity consumption and the cost of fossil fuels, the demand for renewable energy generation is growing. Wind power, as a major renewable energy source, is attracting increasing attention due to its clean, pollution-free, renewable, and high energy content. However, wind power is subject to fluctuations in wind power, resulting in unstable output power, making it difficult to determine the price of wind power. Wind power refers to the power generated by a wind turbine. Furthermore, sudden changes in wind power can lead to a sudden increase in wind turbine load due to the lag in wind turbine control systems, potentially damaging the turbine.

[0003] To optimize wind power pricing strategies and protect wind turbine safety, deep learning models are currently trained using collected wind turbine data. These models are then used to predict wind power, which is then used to optimize wind power pricing and adaptively control wind turbines. However, this approach only considers wind turbine data and ignores the impact of other factors on wind power, resulting in poor accuracy. Summary of the Invention

[0004] The problem solved by the present invention is how to improve the accuracy of wind power prediction.

[0005] To solve the above problems, the present invention provides a wind power prediction model training method, a wind power prediction method, a device and a medium.

[0006] In a first aspect, the present invention provides a wind power prediction model training method, comprising:

[0007] Acquire time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time;

[0008] Performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0009] fusing the first feature data and the second feature data to obtain training data;

[0010] The training data is used to train a pre-built deep neural network to obtain a wind power prediction model.

[0011] Optionally, the wind turbine data includes at least one of average wind speed, rotor speed and wind turbine status, and the meteorological numerical data includes at least one of horizontal and vertical wind speeds, temperature and humidity;

[0012] The method of obtaining the time-series wind turbine data of the wind turbine and the meteorological numerical data of the wind farm where the wind turbine is located includes: obtaining the labeled wind turbine data of the wind turbine at different times, and the meteorological numerical data within an N*N grid centered on the wind farm, and generating the labeled time-series wind turbine data based on the labeled wind turbine data at each time, wherein the label is wind power, and N is greater than or equal to 1.

[0013] Optionally, before extracting features from the time-series wind turbine data based on the self-attention mechanism, the method further includes:

[0014] Sequentially dividing the time series wind turbine data according to a preset time span to obtain a plurality of wind turbine data sequences;

[0015] An average value of the fan data in each of the fan data sequences is determined, and the average values ​​corresponding to each of the fan data sequences are sequentially combined to obtain processed time-series fan data.

[0016] Optionally, the extracting features from the time-series wind turbine data based on the self-attention mechanism includes:

[0017] Obtaining the relative position of each wind turbine data in the time-series wind turbine data, time information of each wind turbine data, and wind turbine status information at different times;

[0018] Determine a position code of each of the wind turbine data according to the relative position, determine a time sequence code of each of the wind turbine data according to the time information, and determine a state code corresponding to each of the wind turbine data according to the wind turbine state information;

[0019] Converting the wind turbine data, the position code, the time series code, and the state code to the same dimension and accumulating them to obtain an input data sequence;

[0020] Feature extraction is performed on the input data sequence based on a multi-head self-attention mechanism to obtain the first feature data.

[0021] Optionally, the input data sequence includes multiple input data, and the extracting features from the input data sequence based on the multi-head self-attention mechanism includes:

[0022] For one of the input data, feature extraction is performed respectively according to a plurality of preset weight pairs to obtain a plurality of sub-feature data;

[0023] Concatenating all the sub-feature data of the input data to obtain a feature vector;

[0024] Performing a linear transformation on the feature vector to obtain the first feature data corresponding to the input data.

[0025] In a second aspect, the present invention provides a wind power prediction method, comprising:

[0026] Acquiring time-series wind turbine data of a target wind turbine and meteorological numerical data of a wind farm where the target wind turbine is located;

[0027] Performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0028] fusing the first feature data and the second feature data to obtain input data;

[0029] Inputting the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine;

[0030] Wherein, the wind power prediction model adopts the wind power prediction model training method as described in any one of the first aspects.

[0031] In a third aspect, the present invention provides a wind power prediction model training device, comprising:

[0032] A first acquisition module is configured to acquire time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time;

[0033] a first extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0034] a first fusion module, configured to fuse the first feature data and the second feature data to obtain training data;

[0035] The training module is used to train a pre-built deep neural network using the training data to obtain a wind power prediction model.

[0036] In a fourth aspect, the present invention provides a wind power prediction device, comprising:

[0037] A second acquisition module is used to acquire the time-series wind turbine data of the target wind turbine and the meteorological numerical data of the wind farm where the target wind turbine is located;

[0038] a second extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0039] a second fusion module, configured to fuse the first feature data and the second feature data to obtain input data;

[0040] A prediction module, configured to input the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine;

[0041] Wherein, the wind power prediction model adopts the wind power prediction model training method as described in any one of the first aspects.

[0042] In a fifth aspect, the present invention provides an electronic device comprising a memory and a processor;

[0043] The memory is used to store computer programs;

[0044] The processor is configured to implement the wind power prediction model training method as described in any one of the first aspects or the wind power prediction method as described in the second aspect when executing the computer program.

[0045] In a sixth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the wind power prediction model training method as described in any one of the first aspects or the wind power prediction method as described in the second aspect is implemented.

[0046] The beneficial effects of the wind power prediction model training method, wind power prediction method, device and medium of the present invention are as follows: obtaining the time series wind turbine data of the wind turbine and the meteorological numerical data of the wind farm where the wind turbine is located, extracting the first feature data from the time series wind turbine data based on the self-attention mechanism, being able to fully learn the historical time series wind turbine data, filtering out noise, improving the accuracy of the extracted first feature data, and thereby improving the accuracy of the wind power prediction model obtained through subsequent training. Extracting the second feature data from the meteorological numerical data based on the cross-attention mechanism can prevent data overflow and improve computing efficiency by scaling the data, and can effectively process meteorological data of larger dimensions. The first feature data and the second feature data are integrated to obtain training data, and the training data is used to train a pre-built deep neural network to obtain a wind power prediction model. The training data not only considers the wind turbine data, but also integrates the meteorological numerical data that affects the wind power, thereby improving the comprehensiveness of the training data, and thereby improving the accuracy of the wind power prediction model obtained through training. The wind power prediction model can accurately predict the wind power within a certain period of time in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic flow chart of a wind power prediction model training method according to an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of the process of extracting features based on the self-attention mechanism according to an embodiment of the present invention;

[0049] Figure 3 1 is a flow chart of a wind power prediction method according to another embodiment of the present invention;

[0050] Figure 4 This is a structural diagram of a wind power prediction model training device according to another embodiment of the present invention;

[0051] Figure 5 The figure is a structural diagram of a wind power prediction device according to another embodiment of the present invention. DETAILED DESCRIPTION

[0052] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0053] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0054] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0055] It should be noted that the modifications of "one" and "multiple" mentioned in the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0056] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0057] like Figure 1 As shown, an embodiment of the present invention provides a wind power prediction model training method, comprising:

[0058] Step S110 , obtaining time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time.

[0059] Specifically, the wind turbine data includes at least one of average wind speed, rotor speed, and wind turbine status. The meteorological numerical data includes at least one of horizontal and vertical wind speeds, temperature, and humidity. The wind turbine data collected at different times by a data acquisition and monitoring system can be sorted by time to obtain time-series wind turbine data. The numerical meteorological data can be obtained from a meteorological center. The time-series wind turbine data can be tagged with the corresponding wind power.

[0060] In step S120 , feature extraction is performed on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and feature extraction is performed on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data.

[0061] Specifically, time-series wind turbine data consists of W*F data, where W represents the number of time points and F represents the amount of data at each time point. This data can be converted into 1*D1-dimensional input data using flattening and linear layers, and feature extraction is performed using a self-attention mechanism. Meteorological numerical data consists of N*N*G data, where N is the grid size centered on the wind turbine and is greater than or equal to 1, and G is the feature dimension of the numerical meteorological data. This data can be convolved into 1*D2-dimensional input data using a series of 3*3 convolution kernels, and feature extraction is performed using a cross-attention mechanism. Using a small 3*3 convolution kernel can more accurately extract the nonlinear features of numerical meteorological data than using a large convolution kernel. The cross-attention mechanism can use a scaled point multiplication method to improve computational efficiency, prevent data overflow, and effectively enhance computational performance for larger dimensional vectors.

[0062] Step S130: fusing the first feature data and the second feature data to obtain training data.

[0063] Specifically, we apply the spatial properties of wind, namely, the wind speed at the next time step is related to the wind speed at the current adjacent grid points. For example, if the wind blows from the current grid point to the adjacent grid point, the current wind speed is related to the current grid point, and the wind speed at the next time step is related to the adjacent grid point. We perform spatiotemporal fusion on the first characteristic data of the time-series wind turbine data and the second characteristic data of the meteorological numerical data, using a method of exchanging space for time to obtain training data.

[0064] Step S140: Using the training data to train a pre-built deep neural network to obtain a wind power prediction model.

[0065] Specifically, the training data can be divided into a training set and a test set. The training set is used to train the constructed deep neural network to obtain the trained deep neural network, i.e., the wind power prediction model. The test set can then be used to evaluate the prediction accuracy of the wind power prediction model.

[0066] In this embodiment, the time series wind turbine data of the wind turbine and the meteorological numerical data of the wind farm where the wind turbine is located are obtained, and the first feature data is extracted from the time series wind turbine data based on the self-attention mechanism. This can fully learn the historical time series wind turbine data, filter out noise, and improve the accuracy of the extracted first feature data, thereby improving the accuracy of the wind power prediction model obtained through subsequent training. The second feature data is extracted from the meteorological numerical data based on the cross-attention mechanism. By scaling the data, data overflow can be prevented, computing efficiency can be improved, and meteorological data of larger dimensions can be effectively processed. The first feature data and the second feature data are integrated to obtain training data, and the training data is used to train a pre-built deep neural network to obtain a wind power prediction model. The training data not only considers the wind turbine data, but also integrates the meteorological numerical data that affects wind power, thereby improving the comprehensiveness of the training data and thereby improving the accuracy of the wind power prediction model obtained through training. The wind power prediction model can make ultra-short-term predictions of wind power, for example, predict the wind power within 4 hours after the reporting point.

[0067] Optionally, obtaining the time-series wind turbine data of the wind turbine and the meteorological numerical data of the wind farm where the wind turbine is located includes: obtaining the labeled wind turbine data of the wind turbine at different times, and the meteorological numerical data within an N*N grid centered on the wind farm, and generating the labeled time-series wind turbine data based on the labeled wind turbine data at each time, wherein the label is wind power, and N is greater than or equal to 1.

[0068] Specifically, labeled wind turbine data for wind turbines at different times can be obtained and sorted in chronological order to generate labeled time-series wind turbine data. For meteorological numerical data, the meteorological numerical data downloaded from the meteorological center is in nc format. The Python library functions provided by the meteorological center can be used to read the data by coordinate and category. The read data is then reorganized into time-sorted tabular data, which is then organized into N*N*G three-dimensional grid data, where N is the grid size, preferably 7, and each grid width is 0.125 degrees. G is the number of features.

[0069] It should be noted that existing techniques typically use interpolation to fuse the two types of data. For example, interpolation refines the temporal granularity of meteorological data, aligning them with wind turbine data at a specific time step. The aligned meteorological and wind turbine data are then spliced ​​and fused. This method is simple and easy to implement, but the interpolated meteorological data loses its volatility. The interpolated meteorological data, such as wind speed, differs significantly from the actual data, introducing noise that negatively impacts the prediction performance of the trained wind power prediction model.

[0070] In this optional embodiment, compared with the prior art method of data refinement through interpolation, the spatial attributes of wind are applied, that is, the grid points adjacent to the current grid point in the next time step are related. Therefore, the method of exchanging space for time is adopted to refine the meteorological numerical data at a spatial granularity, and then the meteorological numerical data and the time-series wind turbine data are spatially and temporally fused, which solves the problem of losing data volatility and introducing noise data caused by data refinement through interpolation, can improve the training effect of the deep model, and improve the prediction accuracy of the wind power prediction model obtained by training.

[0071] Optionally, before extracting features from the time-series wind turbine data based on the self-attention mechanism, the method further includes:

[0072] The time-series wind turbine data is divided in sequence according to a preset time span to obtain a plurality of wind turbine data sequences.

[0073] For example, assuming that the time span is 10 minutes, the wind turbine data in the time-series wind turbine data are sequentially intercepted to obtain a plurality of wind turbine data sequences, wherein each wind turbine data sequence includes 10 minutes of wind turbine data.

[0074] An average value of the fan data in each of the fan data sequences is determined, and the average values ​​corresponding to each of the fan data sequences are sequentially combined to obtain processed time-series fan data.

[0075] Specifically, the fan data in each fan data sequence are averaged to obtain the average value corresponding to each fan data sequence, and all average values ​​are combined according to the chronological order to obtain the processed time series fan data as the input data for feature extraction.

[0076] In this optional embodiment, the average value can reflect the fan speed within a time period. Compared with selecting the fan data at a specific time point as the input data for feature extraction, it can avoid the impact of wind fluctuations on the data, thereby improving the accuracy of the extracted features. At the same time, there is no need to fill in the data of the missing time period to avoid introducing data noise and affecting the accuracy of feature extraction.

[0077] Optionally, the extracting features from the time-series wind turbine data based on the self-attention mechanism includes:

[0078] The relative position of each wind turbine data in the time-series wind turbine data, the time information of each wind turbine data and the wind turbine status information at different times are obtained.

[0079] The position code of each of the wind turbine data is determined according to the relative position, the time sequence code of each of the wind turbine data is determined according to the time information, and the state code corresponding to each of the wind turbine data is determined according to the wind turbine state information.

[0080] Specifically, the wind turbine data may be encoded to obtain a characteristic code. The time sequence code may include a month code, a date code, and an hour code, and the status code may include a power-limited operation code and a wind turbine status code.

[0081] The wind turbine data, the position code, the time sequence code, and the state code are converted into the same dimension and accumulated to obtain an input data sequence.

[0082] Feature extraction is performed on the input data sequence based on a multi-head self-attention mechanism to obtain the first feature data.

[0083] It should be noted that for two segments of time-series wind turbine data that are in different seasons and times, or the wind turbines are in different power-limited operating states, the feature extraction based on the self-attention mechanism may not be able to distinguish the difference between the two, resulting in the same feature data extracted from the time-series wind turbine data at both ends, causing wind power prediction deviation.

[0084] In this optional embodiment, location information, state information and time information are integrated into the wind turbine data so that the input data obtained after the information integration can be distinguished from each other, avoiding the problem of similar time series wind turbine data at different times and states being difficult to distinguish when extracting time series features based on the self-attention mechanism, thereby improving the accuracy of wind power prediction.

[0085] Optionally, the input data sequence includes multiple input data, and the extracting features from the input data sequence based on the multi-head self-attention mechanism includes:

[0086] For one of the input data, feature extraction is performed respectively according to a plurality of preset weight pairs to obtain a plurality of sub-feature data;

[0087] Concatenating all the sub-feature data of the input data to obtain a feature vector;

[0088] Performing a linear transformation on the feature vector to obtain the first feature data corresponding to the input data.

[0089] Specifically, if Figure 2 As shown, a set of weight pairs (Wq, Wk, Wv) are used to extract features from the input data sequence Xt through a self-attention mechanism to obtain a sub-feature sequence Xt'. The sub-feature sequence Xt' includes multiple sub-feature data, and the sub-feature data corresponds to the input data. Feature extraction is performed on the input data sequence using multiple sets of weight pairs to obtain multiple sub-feature sequences. The corresponding sub-feature data in each sub-feature sequence are spliced ​​to obtain the feature vector corresponding to each sub-feature data. For example, the first sub-feature data in the first sub-feature sequence is spliced ​​with the first sub-feature data in the second sub-feature sequence, and so on until the first sub-feature data in the Nth sub-feature sequence to obtain a feature vector corresponding to a sub-feature data. The feature vector can be converted into first feature data through linear transformation, and the first feature data corresponding to each sub-feature data can be combined to obtain a first feature data sequence. The specific process of feature extraction using the self-attention mechanism is prior art and will not be repeated here.

[0090] In this optional embodiment, feature extraction is performed through a multi-head self-attention mechanism. Compared with feature extraction performed through an ordinary self-attention mechanism, data can be learned from multiple angles to avoid deviations, and the accuracy of the extracted features can be improved, thereby improving the effect of model training and improving the accuracy of wind power prediction.

[0091] like Figure 3 As shown, another embodiment of the present invention provides a wind power prediction method, comprising:

[0092] Step S210, acquiring time-series wind turbine data of a target wind turbine and meteorological numerical data of a wind farm where the target wind turbine is located;

[0093] Step S220, performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0094] Step S230, fusing the first feature data and the second feature data to obtain input data;

[0095] Step S240: inputting the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine;

[0096] Wherein, the wind power prediction model adopts the wind power prediction model training method as described above.

[0097] In this embodiment, the time-series wind turbine data of the target wind turbine and the meteorological numerical data of the wind farm where the target wind turbine is located are obtained, and the first feature data can be accurately extracted from the time-series wind turbine data based on the self-attention mechanism, thereby improving the accuracy of wind power prediction. Based on the cross-attention mechanism, the second feature data can be effectively extracted from the meteorological numerical data, and data overflow can be prevented, thereby improving computing efficiency. By fusing the first feature data and the second feature data, the input data not only considers the time-series wind turbine data, but also incorporates the meteorological numerical data that affects wind power, thereby improving the comprehensiveness of considering factors affecting wind power. By inputting the input data into the wind power prediction model trained by the wind power prediction model training method described above, the wind power of the target wind turbine can be predicted in the ultra-short term, for example, the wind power within 4 hours after the reporting point can be predicted with high accuracy.

[0098] like Figure 4 As shown, another embodiment of the present invention provides a wind power prediction model training device, comprising:

[0099] A first acquisition module is configured to acquire time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time;

[0100] a first extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0101] a first fusion module, configured to fuse the first feature data and the second feature data to obtain training data;

[0102] The training module is used to train a pre-built deep neural network using the training data to obtain a wind power prediction model.

[0103] The wind power prediction model training device is used to implement the wind power prediction model training method described above to obtain corresponding beneficial effects.

[0104] like Figure 5As shown, another embodiment of the present invention provides a wind power prediction device, comprising:

[0105] A second acquisition module is used to acquire the time-series wind turbine data of the target wind turbine and the meteorological numerical data of the wind farm where the target wind turbine is located;

[0106] a second extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data;

[0107] a second fusion module, configured to fuse the first feature data and the second feature data to obtain input data;

[0108] A prediction module, configured to input the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine;

[0109] Wherein, the wind power prediction model adopts the wind power prediction model training method as described above.

[0110] The wind power prediction device is used to implement the wind power prediction method described above to obtain corresponding beneficial effects.

[0111] Another embodiment of the present invention provides an electronic device including a memory and a processor; the memory is used to store a computer program; the processor is used to implement the wind power prediction model training method or the wind power prediction method as described above when executing the computer program.

[0112] Yet another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the wind power prediction model training method or the wind power prediction method described above is implemented.

[0113] An electronic device that can serve as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic devices are intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0114] An electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). Various programs and data required for device operation can also be stored in the RAM. The computing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0115] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

[0116] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM). In this application, the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present invention. In addition, the functional units in the various embodiments of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into a single unit. The above-mentioned integrated units can be implemented in the form of hardware or software functional units.

[0117] Although the present invention is disclosed as above, the scope of protection disclosed by the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A wind power prediction model training method, characterized in that: include: Acquire time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time; Performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data; fusing the first feature data and the second feature data to obtain training data; Using the training data to train a pre-built deep neural network to obtain a wind power prediction model; The feature extraction of the time-series wind turbine data based on the self-attention mechanism includes: obtaining the relative position of each wind turbine data in the time-series wind turbine data, the time information of each wind turbine data, and the wind turbine status information at different times; determining the position code of each wind turbine data according to the relative position, determining the time series code of each wind turbine data according to the time information, and determining the state code corresponding to each wind turbine data according to the wind turbine status information; converting the wind data, the position code, the time series code, and the state code to the same dimension and accumulating them to obtain an input data sequence; performing feature extraction on the input data sequence based on the multi-head self-attention mechanism to obtain the first feature data; The input data sequence includes multiple input data, and the feature extraction of the input data sequence based on the multi-head self-attention mechanism includes: for one input data, performing feature extraction on each of the input data according to multiple sets of preset weights to obtain multiple sub-feature data; splicing all the sub-feature data of the input data to obtain a feature vector; and performing linear transformation on the feature vector to obtain the first feature data corresponding to the input data.

2. The wind power prediction model training method according to claim 1, characterized in that: The wind turbine data includes at least one of average wind speed, rotor speed and wind turbine status; the meteorological numerical data includes at least one of horizontal and vertical wind speeds, temperature and humidity; The method of obtaining the time-series wind turbine data of the wind turbine and the meteorological numerical data of the wind farm where the wind turbine is located includes: obtaining the labeled wind turbine data of the wind turbine at different times, and the meteorological numerical data within an N*N grid centered on the wind farm, and generating the labeled time-series wind turbine data based on the labeled wind turbine data at each time, wherein the label is wind power, and N is greater than or equal to 1.

3. The wind power prediction model training method according to claim 2, characterized in that: Before extracting features from the time series wind turbine data based on the self-attention mechanism, the method further includes: Sequentially dividing the time series wind turbine data according to a preset time span to obtain a plurality of wind turbine data sequences; An average value of the fan data in each of the fan data sequences is determined, and the average values ​​corresponding to each of the fan data sequences are sequentially combined to obtain processed time-series fan data.

4. A wind power prediction method, characterized in that: include: Acquiring time-series wind turbine data of a target wind turbine and meteorological numerical data of a wind farm where the target wind turbine is located; Performing feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and performing feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data; fusing the first feature data and the second feature data to obtain input data; Inputting the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine; Wherein, the wind power prediction model adopts the wind power prediction model training method according to any one of claims 1 to 3.

5. A wind power prediction model training device, characterized in that: include: A first acquisition module is configured to acquire time-series wind turbine data of a wind turbine and meteorological numerical data of a wind farm where the wind turbine is located, wherein the meteorological numerical data represents meteorological data of the wind farm, and the time-series wind turbine data represents wind turbine data sorted by time; a first extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data; The feature extraction of the time-series wind turbine data based on the self-attention mechanism includes: obtaining the relative position of each wind turbine data in the time-series wind turbine data, the time information of each wind turbine data and the wind turbine status information at different times; determining the position code of each wind turbine data according to the relative position, determining the time code of each wind turbine data according to the time information, and determining the state code corresponding to each wind turbine data according to the wind status information; converting the wind data, the position code, the time code and the state code to the same dimension, and accumulating them to obtain an input data sequence; performing feature extraction on the input data sequence based on the multi-head self-attention mechanism to obtain the first feature data; the input data sequence includes multiple input data, and the feature extraction of the input data sequence based on the multi-head self-attention mechanism includes: for one input data, performing feature extraction on each of the input data according to multiple sets of preset weights to obtain multiple sub-feature data; splicing all the sub-feature data of the input data to obtain a feature vector; performing linear transformation on the feature vector to obtain the first feature data corresponding to the input data; a first fusion module, configured to fuse the first feature data and the second feature data to obtain training data; The training module is used to train a pre-built deep neural network using the training data to obtain a wind power prediction model.

6. A wind power prediction device, characterized in that: include: A second acquisition module is used to acquire the time-series wind turbine data of the target wind turbine and the meteorological numerical data of the wind farm where the target wind turbine is located; a second extraction module, configured to perform feature extraction on the time-series wind turbine data based on a self-attention mechanism to obtain first feature data, and to perform feature extraction on the meteorological numerical data based on a cross-attention mechanism to obtain second feature data; a second fusion module, configured to fuse the first feature data and the second feature data to obtain input data; A prediction module, configured to input the input data into a wind power prediction model to obtain the wind power corresponding to the target wind turbine; Wherein, the wind power prediction model adopts the wind power prediction model training method according to any one of claims 1 to 3.

7. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the wind power prediction model training method according to any one of claims 1 to 3 or the wind power prediction method according to claim 4 when executing the computer program.

8. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the wind power prediction model training method according to any one of claims 1 to 3 or the wind power prediction method according to claim 4 is implemented.

Images