Three-dimensional wind field prediction method, device, electronic equipment and storage medium

Abstract

The present application provides a three-dimensional wind field prediction method, device, electronic device and storage medium, which relate to the field of wind field prediction. The three-dimensional wind field prediction method includes: obtaining first data information and second data information; performing differential processing on the data in the second data information to obtain first differential data information; performing feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information to obtain a target vector in the latent space; obtaining first atmospheric field data, inputting the first atmospheric field data and the target vector into a target decoder for training learning and spatial conversion, and obtaining three-dimensional wind field prediction information. The present application effectively focuses on the time-series-based change characteristics of the three-dimensional wind field, and also combines a preset large meteorological model to predict the first atmospheric field data, conducts comprehensive analysis and learning on the first atmospheric field data and the target vector, and then outputs accurate three-dimensional wind field prediction information, and is capable of large-scale forecasting.

Description

Technical Field

[0001] The present application relates to the technical field of wind field prediction, and in particular to a three-dimensional wind field prediction method, device, electronic device and storage medium. Background Art

[0002] A three-dimensional wind field is a wind field described in three spatial dimensions and is commonly used for wind energy development and meteorological forecasting. The study of three-dimensional wind fields involves the spatial variations of wind speed, wind direction, and wind pressure. This is achieved through computational fluid dynamics (CFD) models and atmospheric simulations. These models can capture the complexity of wind fields, including factors such as topographic influences, temperature gradients, and atmospheric pressure differences; three-dimensional wind field models are crucial for accurately predicting wind resources and evaluating wind farm performance. With the advancement of computing technology, the accuracy and complexity of three-dimensional wind field models have been significantly improved.

[0003] In related technologies, a new method for small-scale three-dimensional wind field prediction has been proposed, combining the mesoscale weather research and forecasting (WRF) model with computational fluid dynamics (CFD). This hybrid neural network prediction module and CFD module enable accurate simulation and short-term prediction of small-scale three-dimensional wind fields. However, this method is limited to small-scale three-dimensional wind fields and cannot perform large-scale predictions. Furthermore, since predictions are based on the wind field itself, it is difficult to consider the changing patterns of the wind field. Summary of the Invention

[0004] In response to the deficiencies in the prior art, the present application provides a three-dimensional wind field prediction method, device, electronic device and storage medium, which solve the problems of limited three-dimensional wind field prediction process and inaccurate prediction results.

[0005] To achieve the above objectives, this application is implemented through the following technical solutions:

[0006] In the first aspect, an embodiment of the present application provides a three-dimensional wind field prediction method, which includes: obtaining first data information and second data information that characterize the characteristics of the three-dimensional wind field and correspond to different historical moments; based on the time sequence corresponding to the second data information, performing differential processing on the data in the second data information one by one to obtain first differential data information that characterizes the changing state of the three-dimensional wind field; performing feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information based on the first encoder and the second encoder obtained by pre-training, respectively, to obtain a target vector in the latent space; obtaining first atmospheric field data predicted by a preset large meteorological model, inputting the first atmospheric field data and the target vector into a target decoder for training learning and spatial conversion, and obtaining three-dimensional wind field prediction information corresponding to the future moment.

[0007] According to the first aspect of an embodiment of the present application, the first data information corresponds to different historical moments and includes first satellite image data, first ERA5 reanalysis data and first wind field data; the first differential data information corresponds to different historical moments and includes second satellite image data, second ERA5 reanalysis data and second wind field data after time series difference processing.

[0008] According to the first aspect of the embodiment of the present application, the aforementioned first encoder and the second encoder obtained by pre-training respectively perform feature extraction, cross-attention calculation, matching screening and splicing and fusion on the first data information and the first differential data information to obtain a target vector in the latent space, which may specifically include the following steps: based on the first encoder obtained by pre-training and the first discrete codebook obtained by pre-screening, the first satellite image data, the first ERA5 reanalysis data and the first wind field data are processed through a preset first target analysis strategy to obtain a quantized first latent space feature vector; based on the second encoder obtained by pre-training and the second discrete codebook obtained by pre-screening, the second satellite image data, the second ERA5 reanalysis data and the second wind field data are processed through the first target analysis strategy to obtain a quantized second latent space feature vector; the first latent space feature vector and the second latent space feature vector are spliced ​​and fused to generate a target vector in the latent space.

[0009] According to a first aspect of an embodiment of the present application, the aforementioned first encoder obtained by pre-training and the first discrete codebook obtained by pre-screening are used to process the first satellite image data, the first ERA5 reanalysis data, and the first wind field data using a preset first target analysis strategy to obtain a quantized first latent space feature vector. Specifically, the following steps may be included:

[0010] The first wind field data is input into the pre-trained first encoder to generate the first target wind field data represented in the corresponding latent space; the first satellite image data and the first ERA5 reanalysis data are feature extracted by the feature graph pyramid network (FPN) to obtain first multi-scale spatial feature information; a cross-attention calculation is performed on the first multi-scale spatial feature information and the first target wind field data to generate a first initial vector; the first initial vector is matched with the first discrete codebook to obtain a quantized first latent space feature vector.

[0011] According to the first aspect of the embodiment of the present application, before performing feature extraction, cross-attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information based on the pre-trained first encoder and the second encoder, respectively, to obtain a target vector in the latent space, the three-dimensional wind field prediction method may further include the following steps:

[0012] Acquire third data information of the three-dimensional wind field, where the third data information corresponds to different historical moments and includes third satellite image data, third ERA5 reanalysis data, and third wind field data; perform feature extraction on the third satellite image data and the third ERA5 reanalysis data through a feature graph pyramid network (FPN) to obtain second multi-scale spatial feature information; obtain a first encoder based on the third wind field data and the second multi-scale spatial feature information through training using a preset second target analysis strategy, and determine a quantized first eigenvector; acquire second atmospheric field data generated by simulation using a preset large meteorological model; input the second atmospheric field data and the first eigenvector into a first decoder based on a feedforward neural network for processing to perform spatial conversion and generate current wind field data.

[0013] According to the first aspect of the embodiment of the present application, the aforementioned training of obtaining a first encoder based on the third wind field data and the second multi-scale spatial feature information through a preset second target analysis strategy and determining a quantized first feature vector may specifically include the following steps:

[0014] The third wind field data is encoded and trained to generate the second target wind field data represented in the corresponding latent space and obtain the first encoder; the second multi-scale spatial feature information and the second target wind field data are cross-attention calculated to generate a second initial vector; the second initial vector is matched and screened with a preset learnable codebook set to determine the screened first discrete codebook, and the individual in the learnable codebook set that is closest to the second initial vector is determined as the quantized first feature vector.

[0015] According to the first aspect of the embodiment of the present application, before the first encoder and the second encoder obtained by pre-training respectively perform feature extraction, cross attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information to obtain the target vector in the latent space, the three-dimensional wind field prediction method may further include the following steps:

[0016] Acquire fourth data information corresponding to different historical moments of the three-dimensional wind field; perform differential processing on the data in the fourth data information based on the time sequence to obtain second differential data information; the second differential data information includes fourth satellite image data, fourth ERA5 reanalysis data and fourth wind field data; perform feature extraction on the fourth satellite image data and the fourth ERA5 reanalysis data to obtain third multi-scale spatial feature information; based on the fourth wind field data and the third multi-scale spatial feature information, obtain a second encoder through training with a second target analysis strategy to obtain a second discrete codebook and a quantized second eigenvector; obtain third atmospheric field data generated by the preset large meteorological model prediction, perform differential processing on the third atmospheric field data to obtain third differential data information; input the third differential data information and the second eigenvector into a second decoder based on a feedforward neural network for processing to obtain predicted differential wind field data.

[0017] In the second aspect, an embodiment of the present application provides a three-dimensional wind field prediction device, which includes an acquisition module, a differential processing module, a processing module and a training module; wherein the acquisition module is used to acquire first data information and second data information that characterize the characteristics of the three-dimensional wind field and correspond to different historical moments; the differential processing module is used to perform differential processing on the data in the second data information one by one based on the time sequence corresponding to the second data information, and obtain first differential data information that characterizes the changing state of the three-dimensional wind field; the processing module is used to perform feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information based on the first encoder and the second encoder obtained in pre-training, respectively, to obtain a target vector in the latent space; the training module is used to obtain the first atmospheric field data predicted by a preset large meteorological model, input the first atmospheric field data and the target vector into the target decoder for training learning and spatial conversion, and obtain the three-dimensional wind field prediction information corresponding to the future moment.

[0018] In a third aspect, an embodiment of the present application provides an electronic device comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the three-dimensional wind field prediction method in the first aspect is implemented.

[0019] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a program or instruction. When the program or instruction is executed by a processor, the three-dimensional wind field prediction method in the first aspect is implemented.

[0020] This application provides a three-dimensional wind field prediction method, device, electronic device, and storage medium. Compared with the existing technology, it has the following advantages:

[0021] In order to focus on the time-series-based changing characteristics of the three-dimensional wind field, the present application performs differential processing on the data in the acquired second data information, and the obtained first differential data information includes multiple differential data corresponding to different moments; when analyzing the first data information and the first differential data information, pre-training is performed to obtain a first encoder and a second encoder corresponding to the internal characteristics and changing characteristics of the three-dimensional wind field, respectively. The first encoder and the second encoder perform feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information to obtain a target vector in the latent space to fuse the wind field's own characteristics and changing characteristics; at the same time, in order to improve the accuracy of the prediction, the present application also combines a preset large meteorological model to predict the first atmospheric field data, and performs comprehensive analysis and learning on the first atmospheric field data and the target vector based on the target decoder, thereby outputting accurate three-dimensional wind field prediction information, and capable of large-scale forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 This is a flow chart of a three-dimensional wind field prediction method provided in an embodiment of the present application;

[0024] Figure 2 is an exemplary analysis diagram of the third data information provided in an embodiment of the present application;

[0025] Figure 3 is an exemplary analysis diagram of the second differential data information provided in an embodiment of the present application;

[0026] Figure 4 is an exemplary analysis diagram of the first data information and the first differential data information provided in an embodiment of the present application;

[0027] Figure 5 This is a schematic structural diagram of a three-dimensional wind field prediction device provided in an embodiment of the present application;

[0028] Figure 6 This is a schematic structural diagram of another three-dimensional wind field prediction device provided in an embodiment of the present application;

[0029] Figure 7 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0031] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0032] The embodiments of the present application solve the problems of limited three-dimensional wind field prediction process and inaccurate prediction results by providing a three-dimensional wind field prediction method, device, electronic device and storage medium.

[0033] The technical solution in the embodiments of the present application is to solve the above technical problems, and the overall idea is as follows:

[0034] A three-dimensional wind field is a wind field described in three spatial dimensions and is commonly used in wind energy development and meteorological forecasting. The study of three-dimensional wind fields involves the spatial variations of wind speed, direction, and pressure. This is achieved through computational fluid dynamics (CFD) models and atmospheric simulations. These models can capture the complexities of wind fields, including factors such as topographic influences, temperature gradients, and atmospheric pressure differences. Three-dimensional wind field models are crucial for accurately predicting wind resources and evaluating wind farm performance.

[0035] The history of research in three-dimensional wind field prediction can be traced back to the early 20th century, when scientists began using computers to simulate atmospheric flows. With the development of computing technology, the accuracy and complexity of three-dimensional wind field models have been significantly improved. For example, in the field of ground-based detection technology for three-dimensional wind fields in the middle and upper atmosphere, a three-dimensional wind field imaging technology for the middle and upper atmosphere based on multi-station meteor radar network observations has been developed using a multi-station meteor radar system, achieving detailed detection of three-dimensional wind fields in the mesosphere-lower thermosphere atmospheric region. In recent years, deep learning methods have demonstrated great advantages in various prediction tasks. They are good at handling nonlinear processes and can use multi-source data to capture underlying physical relationships. The use of big data and machine learning technologies, such as deep neural networks, has further improved the accuracy of three-dimensional wind field predictions.

[0036] In related technologies, a study has proposed a new method for small-scale three-dimensional wind field prediction that combines the mesoscale weather research and forecasting WRF model with computational fluid dynamics (CFD). The method achieves accurate simulation and short-term prediction of small-scale three-dimensional wind fields by hybridizing the neural network prediction module and the CFD module. However, this method is limited to small-scale three-dimensional wind fields and cannot perform large-scale predictions. Related technologies have the following main disadvantages: (1) Previous models all make predictions based on the wind field itself and rarely focus on the changing characteristics of the wind field. (2) Traditional forecasting models consume a lot of computing resources. (3) Most methods are limited to small-scale three-dimensional wind fields and are difficult to perform large-scale predictions. (4) Most artificial intelligence methods have poor robustness, rely on reanalysis data, and cannot perform real-time predictions.

[0037] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0038] The following first introduces a three-dimensional wind field prediction method provided by an embodiment of the present application.

[0039] The present invention provides a flow chart of a three-dimensional wind field prediction method, as shown in FIG. Figure 1 As shown, the three-dimensional wind field prediction method may include the following steps S110-S140.

[0040] S110 , obtaining first data information and second data information that represent three-dimensional wind field characteristics and correspond to different historical moments.

[0041] S120 . Based on the time sequence corresponding to the second data information, perform differential processing on the data in the second data information one by one to obtain first differential data information representing a change state of the three-dimensional wind field.

[0042] S130, performing feature extraction, cross-attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information based on the first encoder and the second encoder obtained by pre-training, respectively, to obtain a target vector in the latent space.

[0043] S140, obtaining the first atmospheric field data predicted by the preset large meteorological model, inputting the first atmospheric field data and the target vector into the target decoder for training learning and spatial conversion, and obtaining the three-dimensional wind field prediction information corresponding to the future moment.

[0044] The above is a specific implementation method of the three-dimensional wind field prediction method provided in the embodiment of the present application. It can be understood that the present application focuses on the time-series-based change characteristics of the three-dimensional wind field, and performs differential processing on the data in the acquired second data information. The first differential data information obtained includes multiple differential data corresponding to different moments; when analyzing the first data information and the first differential data information, the first encoder and the second encoder corresponding to the internal characteristics and change characteristics of the three-dimensional wind field are pre-trained, and the first encoder and the second encoder are used to perform feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information to obtain the target vector in the latent space to fuse the wind field's own characteristics and change characteristics.

[0045] Furthermore, in order to improve the accuracy of the prediction, the present application also combines a preset large meteorological model to predict the first atmospheric field data, conducts comprehensive analysis and learning of the first atmospheric field data and the target vector based on the target decoder, and then outputs accurate three-dimensional wind field prediction information, and is capable of large-scale forecasting.

[0046] In some embodiments, the first data information corresponds to different historical moments and includes first satellite image data, first ERA5 reanalysis data and first wind field data; the first differential data information corresponds to different historical moments and includes second satellite image data, second ERA5 reanalysis data and second wind field data after time series difference processing.

[0047] The first encoder and the second encoder obtained by pre-training respectively perform feature extraction, cross attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information to obtain a target vector in the latent space; that is, the aforementioned S130 may specifically include the following steps:

[0048] S210: Based on the pre-trained first encoder and the pre-screened first discrete codebook, the first satellite image data, the first ERA5 reanalysis data, and the first wind field data are processed using a preset first target analysis strategy to obtain a quantized first latent space feature vector.

[0049] S220. Based on the second encoder obtained by pre-training and the second discrete codebook obtained by pre-screening, the second satellite image data, the second ERA5 reanalysis data, and the second wind field data are processed by the first target analysis strategy to obtain a quantized second latent space feature vector.

[0050] S230: Concatenate and fuse the first latent space feature vector and the second latent space feature vector to generate a target vector in the latent space.

[0051] In an example, see Figure 4 It can be understood that the first data information is the historical data corresponding to real time 1, real time 2, real time 3 and real time 4, and the first data information can characterize the characteristics of the three-dimensional wind farm within a period of history; the first differential data information is the data difference corresponding to real time 1 and real time 0, the data difference corresponding to real time 2 and real time 1, the data difference corresponding to real time 3 and real time 2, and the data difference corresponding to real time 4 and real time 3; for the first data information and the first differential data information, feature extraction is performed by the first encoder and the second encoder obtained by pre-training, respectively, and the extracted features are spliced ​​and fused, so that the present application can pay attention to the characteristics of the three-dimensional wind farm itself and the changing characteristics that may be caused by external influences at the same time.

[0052] In another example, based on the pre-trained first encoder and the pre-screened first discrete codebook, the first satellite image data, the first ERA5 reanalysis data, and the first wind field data are processed using a preset first target analysis strategy to obtain a quantized first latent space feature vector. That is, the aforementioned S210 may specifically include the following steps:

[0053] S310: Input the first wind field data into a pre-trained first encoder to generate first target wind field data represented in a corresponding latent space.

[0054] S320 , performing feature extraction on the first satellite image data and the first ERA5 reanalysis data through a feature graph pyramid network (FPN) to obtain first multi-scale spatial feature information.

[0055] S330: Perform a cross-attention calculation on the first multi-scale spatial feature information and the first target wind field data to generate a first initial vector.

[0056] S340: Match the first initial vector with the first discrete codebook to obtain a quantized first latent space feature vector.

[0057] In some embodiments, before the first encoder and the second encoder obtained by pre-training respectively perform feature extraction, cross-attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information to obtain the target vector in the latent space, that is, before the aforementioned S130, the three-dimensional wind field prediction method may further include:

[0058] S410: Acquire third data information of a three-dimensional wind field, where the third data information corresponds to different historical moments and includes third satellite image data, third ERA5 reanalysis data, and third wind field data.

[0059] S420 , performing feature extraction on the third satellite image data and the third ERA5 reanalysis data through a feature map pyramid network (FPN) to obtain second multi-scale spatial feature information.

[0060] S430 : Based on the third wind field data and the second multi-scale spatial feature information, obtain a first encoder through training using a preset second target analysis strategy, and determine a quantized first feature vector.

[0061] S440: Obtain the second atmospheric field data generated by simulation of a preset large meteorological model.

[0062] S450: Input the second atmospheric field data and the first eigenvector into a first decoder based on a feedforward neural network for processing, so as to perform spatial conversion and generate current wind field data.

[0063] In one example, the aforementioned step S430 may specifically include the following steps:

[0064] S431. Perform encoding training on the third wind field data to generate second target wind field data represented in the corresponding latent space and obtain a first encoder.

[0065] S432: Perform a cross-attention calculation on the second multi-scale spatial feature information and the second target wind field data to generate a second initial vector.

[0066] S433: Match and screen the second initial vector with a preset learnable codebook set to determine a first discrete codebook obtained by the screening, and determine the individual in the learnable codebook set that is closest to the second initial vector as the quantized first eigenvector.

[0067] In the embodiment of the present application, it can be understood that in the aforementioned process of generating the target vector in the latent space, it is necessary to pre-train the first encoder first; to this end, the present application obtains the third data information of the three-dimensional wind field, performs feature extraction on the third satellite image data and the third ERA5 reanalysis data through the feature graph pyramid network FPN, obtains the second multi-scale spatial feature information, and performs encoding training on the third wind field data to generate the second target wind field data in the latent space, and obtains the first encoder through training.

[0068] Furthermore, in the aforementioned matching of the first initial vector with the first discrete codebook to obtain the quantized first latent space feature vector, that is, in the aforementioned S340 process, it is necessary to use the first discrete codebook pre-screened from the learnable codebook set; for this purpose, the present application performs cross-attention calculation on the second multi-scale spatial feature information and the second target wind field data to generate the second initial vector, and then matches and screens the second initial vector with the preset learnable codebook set to determine the first discrete codebook.

[0069] For example, please refer to Figure 2 The third data information obtained by this application corresponds to real time 1, real time 2, real time 3 and real time 4, that is, the third data information corresponds to historical information of 4 different times.

[0070] In some embodiments, before the first encoder and the second encoder obtained by pre-training respectively perform feature extraction, cross-attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information to obtain the target vector in the latent space, that is, before the aforementioned S130, the three-dimensional wind field prediction method may further include:

[0071] S510: Obtain fourth data information of the three-dimensional wind field corresponding to different historical moments.

[0072] S520: Perform differential processing on the data in the fourth data information based on a time sequence to obtain second differential data information; the second differential data information includes fourth satellite image data, fourth ERA5 reanalysis data, and fourth wind field data.

[0073] S530 : Perform feature extraction on the fourth satellite image data and the fourth ERA5 reanalysis data to obtain third multi-scale spatial feature information.

[0074] S540 : Based on the fourth wind field data and the third multi-scale spatial feature information, obtain a second encoder through second target analysis strategy training, obtain a second discrete codebook and a quantized second feature vector.

[0075] S550: Obtain third atmospheric field data generated by the preset large meteorological model prediction, perform differential processing on the third atmospheric field data, and obtain third differential data information.

[0076] S560: Input the third differential data information and the second eigenvector into a second decoder based on a feedforward neural network for processing to obtain predicted differential wind field data.

[0077] In the embodiment of the present application, it can be understood that in the aforementioned process of generating the target vector in the latent space, it is necessary to pre-train the second encoder first. To this end, the present application obtains the fourth data information of the three-dimensional wind field and performs data analysis and processing to obtain the second encoder and the second discrete codebook through the second target analysis strategy training.

[0078] For example, after obtaining the fourth data information corresponding to different historical moments of the three-dimensional wind field, the present application performs differential processing on the data in the fourth data information to obtain the second differential data information; please refer to Figure 3 The second differential data information corresponds to the data difference between real time 1 and real time 0, the data difference between real time 2 and real time 1, the data difference between real time 3 and real time 2, and the data difference between real time 4 and real time 3.

[0079] In some embodiments, the present application provides a three-dimensional wind field prediction device 600, such as Figure 5 As shown, the three-dimensional wind field prediction device 600 may include the following modules:

[0080] An acquisition module 610 is configured to acquire first data information and second data information representing three-dimensional wind field characteristics and corresponding to different historical moments;

[0081] The differential processing module 620 is configured to perform differential processing on the data in the second data information one by one based on the time sequence corresponding to the second data information to obtain first differential data information representing the change state of the three-dimensional wind field;

[0082] A processing module 630 is configured to perform feature extraction, cross-attention calculation, matching screening, and splicing and fusion on the first data information and the first differential data information based on the pre-trained first encoder and the second encoder, respectively, to obtain a target vector in a latent space;

[0083] The training module 640 is used to obtain the first atmospheric field data predicted by the preset meteorological model, input the first atmospheric field data and the target vector into the target decoder for training learning and spatial conversion, and obtain the three-dimensional wind field prediction information corresponding to the future moment.

[0084] According to an embodiment of the present application, any multiple modules among the acquisition module 610, the difference processing module 620, the processing module 630, and the training module 640 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module.

[0085] In some embodiments, the processing module 630 may specifically include:

[0086] A first analysis and quantization unit 631 is configured to process the first satellite image data, the first ERA5 reanalysis data, and the first wind field data using a preset first target analysis strategy based on the pre-trained first encoder and the pre-screened first discrete codebook to obtain a quantized first latent space feature vector;

[0087] A second analysis and quantization unit 632 is configured to process the second satellite image data, the second ERA5 reanalysis data, and the second wind field data using the first target analysis strategy based on the pre-trained second encoder and the pre-screened second discrete codebook to obtain a quantized second latent space feature vector;

[0088] The splicing and fusion unit 633 is used to splice and fuse the first latent space feature vector and the second latent space feature vector to generate a target vector in the latent space.

[0089] In some embodiments, the first analysis and quantification unit 631 may be specifically configured to:

[0090] Inputting the first wind field data into the pre-trained first encoder to generate first target wind field data represented in the corresponding latent space;

[0091] Feature extraction is performed on the first satellite image data and the first ERA5 reanalysis data through a feature graph pyramid network (FPN) to obtain first multi-scale spatial feature information;

[0092] Performing a cross-attention calculation on the first multi-scale spatial feature information and the first target wind field data to generate a first initial vector;

[0093] The first initial vector is matched with the first discrete codebook to obtain a quantized first latent space feature vector.

[0094] In some embodiments, please refer to Figure 6 The three-dimensional wind field prediction device 600 may further include a first pre-training module 650. The first pre-training module 650 may be used to:

[0095] Acquire third data information of the three-dimensional wind field, where the third data information corresponds to different historical moments and includes third satellite image data, third ERA5 reanalysis data, and third wind field data;

[0096] The third satellite image data and the third ERA5 reanalysis data are subjected to feature extraction by the feature graph pyramid network (FPN) to obtain the second multi-scale spatial feature information;

[0097] Based on the third wind field data and the second multi-scale spatial feature information, a first encoder is trained by a preset second target analysis strategy, and a quantized first feature vector is determined;

[0098] Obtain the second atmospheric field data generated by the preset large meteorological model simulation;

[0099] The second atmospheric field data and the first eigenvector are input into a first decoder based on a feedforward neural network for processing, so as to perform spatial conversion and generate current wind field data.

[0100] In some embodiments, the first pre-training module 650 may be specifically used to:

[0101] Performing encoding training on the third wind field data to generate second target wind field data represented in the corresponding latent space and obtain a first encoder;

[0102] Performing cross-attention calculation on the second multi-scale spatial feature information and the second target wind field data to generate a second initial vector;

[0103] The second initial vector is matched and screened with a preset learnable codebook set to determine a first discrete codebook obtained by the screening, and the individual in the learnable codebook set closest to the second initial vector is determined as the quantized first eigenvector.

[0104] In some embodiments, please refer to Figure 6 The three-dimensional wind field prediction device 600 may further include a second pre-training module 660, which may be used to:

[0105] Obtaining fourth data information of the three-dimensional wind field corresponding to different historical moments;

[0106] performing differential processing on the data in the fourth data information based on a time sequence to obtain second differential data information; the second differential data information includes fourth satellite image data, fourth ERA5 reanalysis data, and fourth wind field data;

[0107] performing feature extraction on the fourth satellite image data and the fourth ERA5 reanalysis data to obtain third multi-scale spatial feature information;

[0108] Based on the fourth wind field data and the third multi-scale spatial feature information, a second encoder is obtained by training through a second target analysis strategy to obtain a second discrete codebook and a quantized second feature vector;

[0109] Obtaining third atmospheric field data generated by a preset large meteorological model prediction, performing differential processing on the third atmospheric field data, and obtaining third differential data information;

[0110] The third differential data information and the second eigenvector are input into a second decoder based on a feedforward neural network for processing to obtain predicted differential wind field data.

[0111] Figure 6 Each module in the device shown has the function of implementing each step in the aforementioned three-dimensional wind field prediction method and can achieve its corresponding technical effect. For the sake of brevity, it will not be repeated here.

[0112] In some embodiments, the present application provides an electronic device, the structural diagram of the electronic device is as follows Figure 7 shown.

[0113] The electronic device may include a processor 710 and a memory 720 storing computer program instructions.

[0114] Specifically, the processor 710 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0115] The memory 720 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 720 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 720 may include removable or non-removable (or fixed) media. Where appropriate, the memory 720 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 720 is a non-volatile solid-state memory.

[0116] The memory 720 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Therefore, generally, the memory 720 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it may perform the operations described in any one of the three-dimensional wind field prediction methods in the above-mentioned embodiments.

[0117] The processor 710 implements any one of the three-dimensional wind field prediction methods in the above embodiments by reading and executing computer program instructions stored in the memory 720 .

[0118] In one example, the electronic device may further include a communication interface 730 and a bus 700. Figure 7 As shown, the processor 710 , the memory 720 , and the communication interface 730 are connected via a bus 700 and communicate with each other.

[0119] The communication interface 730 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0120] Bus 700 includes hardware, software or both, and the parts of online data flow metering equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 700 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0121] In addition, in conjunction with the three-dimensional wind field prediction method in the above embodiments, embodiments of the present application may provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the three-dimensional wind field prediction methods in the above embodiments is implemented.

[0122] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0123] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. Programs or code segments can be stored in machine-readable media, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0124] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0125] Aspects of the present disclosure have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or flowchart and the combination of the boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0126] In summary, compared with the prior art, this application has the following beneficial effects:

[0127] 1. In order to focus on the time-series-based change characteristics of the three-dimensional wind field, the present application performs differential processing on the data in the acquired second data information, and the obtained first differential data information includes multiple differential data corresponding to different moments; by performing feature extraction, cross-attention calculation, matching screening and splicing fusion on the first data information and the first differential data information, the target vector in the latent space is obtained, which integrates the wind field's own characteristics and change characteristics.

[0128] 2. In order to improve the accuracy of the prediction, this application combines the preset meteorological large model to predict the first atmospheric field data, and conducts comprehensive analysis and learning on the first atmospheric field data and the target vector based on the target decoder, which can output accurate three-dimensional wind field prediction information.

[0129] 3. This application solves the problem of using and extracting three-dimensional wind field change characteristics, and the entire processing process is low-cost, high-precision, and fast, and can be expanded to large-scale forecast scenarios. This prediction method is not limited to small-scale three-dimensional wind fields.

[0130] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A three-dimensional wind field prediction method, characterized in that: include: Acquire first data information and second data information representing three-dimensional wind field characteristics and corresponding to different historical moments; Based on the time sequence corresponding to the second data information, performing differential processing on the data in the second data information one by one to obtain first differential data information representing a change state of the three-dimensional wind field; Performing feature extraction, cross-attention calculation, matching screening, and splicing fusion on the first data information and the first differential data information based on the pre-trained first encoder and the second encoder, respectively, to obtain a target vector in a latent space; Obtaining first atmospheric field data predicted by a preset large meteorological model, inputting the first atmospheric field data and the target vector into a target decoder for training and spatial conversion, and obtaining three-dimensional wind field prediction information corresponding to the future moment; The first data information corresponds to different historical moments and includes first satellite image data, first ERA5 reanalysis data, and first wind farm data; the first differential data information corresponds to different historical moments and includes second satellite image data, second ERA5 reanalysis data, and second wind farm data processed based on time series difference; The first encoder and the second encoder obtained by pre-training are respectively based on the first data information and the first differential data information to perform feature extraction, cross attention calculation, matching screening, and splicing fusion to obtain a target vector in the latent space, including: Based on the pre-trained first encoder and the pre-screened first discrete codebook, the first satellite image data, the first ERA5 reanalysis data, and the first wind field data are processed using a preset first target analysis strategy to obtain a quantized first latent space feature vector; Based on the pre-trained second encoder and the pre-screened second discrete codebook, the second satellite image data, the second ERA5 reanalysis data, and the second wind field data are processed using the first target analysis strategy to obtain a quantized second latent space feature vector; Concatenate and fuse the first latent space feature vector and the second latent space feature vector to generate a target vector in the latent space; The first encoder obtained by pre-training and the first discrete codebook obtained by pre-screening are used to process the first satellite image data, the first ERA5 reanalysis data, and the first wind field data using a preset first target analysis strategy to obtain a quantized first latent space feature vector, including: Inputting the first wind field data into a pre-trained first encoder to generate first target wind field data represented in a corresponding latent space; Performing feature extraction on the first satellite image data and the first ERA5 reanalysis data through a feature graph pyramid network (FPN) to obtain first multi-scale spatial feature information; Performing a cross-attention calculation on the first multi-scale spatial feature information and the first target wind field data to generate a first initial vector; The first initial vector is matched with the first discrete codebook to obtain a quantized first latent space feature vector.

2. The three-dimensional wind field prediction method according to claim 1, characterized in that: Before performing feature extraction, cross-attention calculation, matching screening, and splicing and fusion on the first data information and the first differential data information based on the first encoder and the second encoder obtained by pre-training to obtain a target vector in a latent space, the three-dimensional wind field prediction method further includes: Acquiring third data information of a three-dimensional wind field, the third data information corresponding to different historical moments and including third satellite image data, third ERA5 reanalysis data, and third wind field data; performing feature extraction on the third satellite image data and the third ERA5 reanalysis data through a feature graph pyramid network (FPN), obtaining second multi-scale spatial feature information; Based on the third wind field data and the second multi-scale spatial feature information, a first encoder is obtained by training through a preset second target analysis strategy, and a quantized first feature vector is determined; Acquire the second atmospheric field data generated by simulation of the preset large meteorological model; The second atmospheric field data and the first eigenvector are input into a first decoder based on a feedforward neural network for processing, so as to perform spatial conversion and generate current wind field data.

3. The three-dimensional wind field prediction method according to claim 2, characterized in that: The method of obtaining a first encoder by training a preset second target analysis strategy based on the third wind field data and the second multi-scale spatial feature information, and determining a quantized first feature vector, includes: Performing encoding training on the third wind field data to generate second target wind field data represented in a corresponding latent space and obtain a first encoder; performing a cross-attention calculation on the second multi-scale spatial feature information and the second target wind field data to generate a second initial vector; The second initial vector is matched and screened with a preset learnable codebook set to determine a first discrete codebook obtained by the screening, and an individual in the learnable codebook set that is closest to the second initial vector is determined as a quantized first eigenvector.

4. The three-dimensional wind field prediction method according to claim 2, characterized in that: Before performing feature extraction, cross-attention calculation, matching screening, and splicing and fusion on the first data information and the first differential data information based on the first encoder and the second encoder obtained by pre-training to obtain a target vector in a latent space, the three-dimensional wind field prediction method further includes: Obtaining fourth data information of the three-dimensional wind field corresponding to different historical moments; performing differential processing on the data in the fourth data information based on a time sequence to obtain second differential data information; the second differential data information includes fourth satellite image data, fourth ERA5 reanalysis data, and fourth wind field data; performing feature extraction on the fourth satellite image data and the fourth ERA5 reanalysis data to obtain third multi-scale spatial feature information; Based on the fourth wind field data and the third multi-scale spatial feature information, a second encoder is trained by the second target analysis strategy to obtain a second discrete codebook and a quantized second feature vector; Acquire third atmospheric field data generated by the preset large meteorological model prediction, perform differential processing on the third atmospheric field data, and obtain third differential data information; The third differential data information and the second eigenvector are input into a second decoder based on a feedforward neural network for processing to obtain predicted differential wind field data.

5. A three-dimensional wind field prediction device, characterized in that: include: An acquisition module, configured to acquire first data information and second data information representing three-dimensional wind field characteristics and corresponding to different historical moments; a differential processing module, configured to perform differential processing on the data in the second data information one by one based on the time sequence corresponding to the second data information, to obtain first differential data information representing a change state of the three-dimensional wind field; a processing module, configured to perform feature extraction, cross-attention calculation, matching screening, and splicing and fusion on the first data information and the first differential data information based on the pre-trained first encoder and the second encoder, respectively, to obtain a target vector in a latent space; A training module is used to obtain first atmospheric field data predicted by a preset large meteorological model, input the first atmospheric field data and the target vector into a target decoder for training learning and spatial conversion, and obtain three-dimensional wind field prediction information corresponding to the future time; The first data information corresponds to different historical moments and includes first satellite image data, first ERA5 reanalysis data, and first wind farm data; the first differential data information corresponds to different historical moments and includes second satellite image data, second ERA5 reanalysis data, and second wind farm data processed based on time series difference; The first encoder and the second encoder obtained by pre-training are respectively based on the first data information and the first differential data information to perform feature extraction, cross attention calculation, matching screening, and splicing fusion to obtain a target vector in the latent space, including: Based on the pre-trained first encoder and the pre-screened first discrete codebook, the first satellite image data, the first ERA5 reanalysis data, and the first wind field data are processed using a preset first target analysis strategy to obtain a quantized first latent space feature vector; Based on the pre-trained second encoder and the pre-screened second discrete codebook, the second satellite image data, the second ERA5 reanalysis data, and the second wind field data are processed using the first target analysis strategy to obtain a quantized second latent space feature vector; Concatenate and fuse the first latent space feature vector and the second latent space feature vector to generate a target vector in the latent space; The first encoder obtained by pre-training and the first discrete codebook obtained by pre-screening are used to process the first satellite image data, the first ERA5 reanalysis data, and the first wind field data using a preset first target analysis strategy to obtain a quantized first latent space feature vector, including: Inputting the first wind field data into a pre-trained first encoder to generate first target wind field data represented in a corresponding latent space; Performing feature extraction on the first satellite image data and the first ERA5 reanalysis data through a feature graph pyramid network (FPN) to obtain first multi-scale spatial feature information; Performing a cross-attention calculation on the first multi-scale spatial feature information and the first target wind field data to generate a first initial vector; The first initial vector is matched with the first discrete codebook to obtain a quantized first latent space feature vector.

6. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the three-dimensional wind field prediction method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the three-dimensional wind field prediction method according to any one of claims 1 to 4 is implemented.

Images