A method, apparatus, equipment and storage medium for predicting wind power output
By dividing wind power generation prediction into two parts—meteorological numerical spatiotemporal prediction and power generation mapping—a deep learning model with an RMT-LSTM structure is adopted, which solves the problem of poor long-term prediction stability in existing technologies and achieves more accurate wind power generation prediction.
Patent Information
- Application Number
- CN202510248097.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-04
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2045-03-04
AI Technical Summary
Existing artificial intelligence methods for wind power generation forecasting suffer from high accuracy in ultra-short-term predictions but poor stability in long-term predictions, making it difficult to meet the requirements for accurate, stable, and long-term forecasts.
The problem of wind power generation prediction is divided into two parts: meteorological numerical spatiotemporal prediction and power generation mapping. A deep learning model with RMT-LSTM structure is adopted. The spatial feature extraction capability is improved by using the RMT model and combined with the ST-LSTM model for prediction.
It improves the accuracy and stability of wind power generation forecasting, enabling accurate forecasting over long periods of time.
Smart Images

Figure CN120109795B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power generation, and in particular to a wind power prediction method, apparatus, equipment, and computer-readable storage medium. Background Technology
[0002] With increasing global focus on sustainable development, the development and utilization of new energy sources, especially wind energy, has become a key approach to addressing climate change and optimizing the energy structure. Accurate forecasting of wind power generation can better optimize power system dispatch and load balancing, improve the operational efficiency and economic benefits of wind farms, thereby promoting the development and application of renewable energy.
[0003] Artificial intelligence methods are based on data-driven approaches, building machine learning or deep learning models for power prediction. This method can effectively model the nonlinear relationships between data and has high prediction accuracy. However, current artificial intelligence methods usually only make ultra-short-term predictions based on time series data, with poor prediction stability and long-term prediction capabilities. Summary of the Invention
[0004] The purpose of this invention is to provide a wind power prediction method, device, equipment and storage medium for application in the field of wind power generation. The method divides the wind power prediction problem into two problems: meteorological numerical spatiotemporal prediction and power generation mapping, and solves them sequentially. In the meteorological numerical spatiotemporal prediction process, an RMT-LSTM structure that fuses the RMT model and the ST-LSTM model is used to improve the model's ability to extract spatial features and make the prediction results more accurate.
[0005] To address the aforementioned technical problems, this invention provides a wind power prediction method, comprising:
[0006] Acquire historical power generation data of the wind farm and historical meteorological data of the target area, and divide the historical meteorological data into multiple time series samples according to time sequence;
[0007] The RMT-LSTM spatiotemporal prediction deep learning network is trained based on the meteorological dataset constructed from the time series samples to obtain the trained meteorological numerical spatiotemporal prediction model.
[0008] A multi-channel spatial deep learning network is trained on a spatial data conversion dataset constructed based on the historical power generation data and the historical meteorological data to obtain a spatial data conversion model for power generation.
[0009] The meteorological data to be measured is input into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and the meteorological element prediction results are input into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0010] Optionally, acquiring historical power generation data of the wind farm and historical meteorological data of the target area includes:
[0011] Obtain the initial historical power generation data of the wind farm and the initial historical meteorological data of the target area;
[0012] Interpolation processing is performed on the missing data in the initial historical power generation data and the initial historical meteorological data to obtain complete historical power generation data and complete historical meteorological data.
[0013] The historical power generation data and the historical meteorological data are obtained by standardizing the supplemented historical power generation data and the supplemented historical meteorological data.
[0014] Optionally, the method further includes:
[0015] The Gaussian weights of each type of data in the historical meteorological data are determined based on the Gaussian weighting function.
[0016] The Gaussian weights are stacked with the historical meteorological data in the channel dimension.
[0017] Optionally, the historical meteorological data includes: historical satellite remote sensing data, historical numerical weather upper-air data, and historical numerical weather surface data.
[0018] Optionally, the spatial data conversion dataset constructed based on the historical power generation data and the historical meteorological data is used to train a multi-channel spatial deep learning network to obtain a power generation spatial data conversion model, including:
[0019] The multi-channel spatial deep learning network is trained using the spatial data conversion dataset constructed based on the historical power generation data and the historical numerical weather surface data in the historical meteorological data to obtain the spatial data conversion model for power generation.
[0020] Optionally, the RMT-LSTM spatiotemporal prediction deep learning network has 4 layers.
[0021] Optionally, the multi-channel spatial deep learning network is a CNN model or a Swing-Transformer model.
[0022] To solve the above-mentioned technical problems, the present invention provides a wind power prediction device, comprising:
[0023] The first module is used to acquire historical power generation data of the wind farm and historical meteorological data of the target area, and to divide the historical meteorological data into multiple time series samples according to the time sequence.
[0024] The second module is used to train the RMT-LSTM spatiotemporal prediction deep learning network based on the meteorological dataset constructed from the time series samples, so as to obtain the trained meteorological numerical spatiotemporal prediction model.
[0025] The third module is used to train a multi-channel spatial deep learning network based on the spatial data conversion dataset constructed from the historical power generation data and the historical meteorological data, so as to obtain the power generation spatial data conversion model.
[0026] The fourth module is used to input the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and to input the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0027] To solve the above-mentioned technical problems, the present invention provides an electronic device, comprising:
[0028] Memory, used to store computer programs;
[0029] A processor is used to implement the wind power prediction method described above when executing the computer program.
[0030] To address the aforementioned technical problems, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the aforementioned wind power prediction method.
[0031] As can be seen, the method of the present invention acquires historical power generation data of wind farms and historical meteorological data of the target area, and divides the historical meteorological data into multiple time series samples according to time sequence; trains the RMT-LSTM spatiotemporal prediction deep learning network based on the meteorological dataset constructed based on the time series samples to obtain the trained meteorological numerical spatiotemporal prediction model; trains a multi-channel spatial deep learning network based on the spatial data conversion dataset constructed based on the historical power generation data and historical meteorological data to obtain the power generation spatial data conversion model; inputs the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and inputs the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0032] This invention proposes a wind power generation prediction method, which divides the wind power generation prediction problem into two problems: meteorological numerical spatiotemporal prediction and power generation mapping, and solves them sequentially. In the meteorological numerical spatiotemporal prediction process, an RMT-LSTM structure that fuses the RMT model and the ST-LSTM model is adopted to improve the model's ability to extract spatial features and make the prediction results more accurate. Attached Figure Description
[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0034] Figure 1 A flowchart of a wind power prediction method provided in an embodiment of the present invention;
[0035] Figure 2 This is an example of an RMT-LSTM spatiotemporal prediction deep learning network architecture provided in an embodiment of the present invention;
[0036] Figure 3 A general training process example diagram of a meteorological numerical spatiotemporal prediction model provided in an embodiment of the present invention;
[0037] Figure 4 A general training process example diagram of a power generation spatial data conversion model provided in an embodiment of the present invention;
[0038] Figure 5 This is a general flowchart illustrating a wind power generation prediction method provided in an embodiment of the present invention.
[0039] Figure 6 This is a structural block diagram of a wind power prediction device provided in an embodiment of the present invention. Detailed Implementation
[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0041] With increasing global focus on sustainable development, the development and utilization of new energy sources, especially wind energy, has become a key approach to addressing climate change and optimizing the energy structure. Accurate forecasting of wind power generation can better optimize power system dispatch and load balancing, improve the operational efficiency and economic benefits of wind farms, thereby promoting the development and application of renewable energy.
[0042] Currently, there are three main methods for predicting wind power output: statistical methods, physical model methods, and artificial intelligence methods. Statistical methods use statistical models to analyze the statistical characteristics of historical data for power prediction. While simple to implement, these methods have weak nonlinear modeling capabilities and struggle to handle complex relationships between meteorological factors and wind power. Physical model methods predict future wind power output by simulating atmospheric and wind field distribution. These methods can effectively simulate the physical mechanisms of actual wind power generation and have high accuracy, but require significant computing resources and have poor real-time prediction performance. Artificial intelligence methods are data-driven, building machine learning or deep learning models for power prediction. These methods can effectively model nonlinear relationships between data and have high prediction accuracy. However, current artificial intelligence methods typically only provide ultra-short-term predictions of a few hours based on time series data, resulting in poor prediction stability and long-term prediction capabilities.
[0043] The three methods mentioned above are insufficient to simultaneously meet the needs of accurate, stable, and long-term wind power prediction under the current development of new energy sources. Therefore, how to achieve these requirements in wind power prediction tasks is a problem that needs to be further solved.
[0044] To meet the need for accurate, stable, and long-term prediction of wind power generation, this invention proposes a wind power generation prediction method based on meteorological numerical spatiotemporal forecasting. The wind power generation prediction problem is divided into two problems: meteorological numerical spatiotemporal forecasting and power generation mapping, which are solved sequentially.
[0045] The following combination Figure 1 , Figure 1 A flowchart of a wind power prediction method provided in an embodiment of the present invention, the method may include:
[0046] S101: Obtain historical power generation data of the wind farm and historical meteorological data of the target area, and divide the historical meteorological data into multiple time series samples according to time sequence.
[0047] This embodiment can acquire historical power generation data of a selected wind farm and historical meteorological data of its target area. This embodiment does not limit the specific type of meteorological data; it can generally include multi-source meteorological data such as satellite remote sensing data, numerical weather prediction (NMR) upper-air data, and numerical weather prediction (NMR) surface data. NMR upper-air data can include: upper-air wind speed, upper-air wind direction, upper-air temperature, and upper-air humidity; NMR surface data can include: surface wind speed, surface wind direction, surface temperature, surface humidity, and surface air pressure. This embodiment can align the acquired data in both spatial and temporal dimensions.
[0048] In this embodiment, after collecting initial historical meteorological data and historical power generation data, the data can be preprocessed to obtain samples that meet the model training specifications.
[0049] This embodiment does not limit the specific method of preprocessing. Generally, the initial historical power generation data of the wind farm and the initial historical meteorological data of the target area can be obtained. The missing data in the initial historical power generation data and the initial historical meteorological data are interpolated to obtain the complete historical power generation data and the complete historical meteorological data. The complete historical power generation data and the initially completed historical meteorological data are standardized to obtain the historical power generation data and the historical meteorological data.
[0050] This embodiment does not limit the specific method of interpolating missing data; generally, bilinear interpolation or Newton interpolation can be selected.
[0051] This embodiment does not limit the specific method of standardization. Normalization and Z-SCORE standardization can generally be selected. Z-SCORE, also known as standard score, is the process of dividing the difference between a number and the mean by the standard deviation.
[0052] Furthermore, using a Gaussian weighting function can amplify the prediction weight of the central location, i.e., the location of the wind farm, allowing the meteorological prediction model to focus on predicting meteorological data changes in the wind farm's location and surrounding area. Specifically, this embodiment can determine the Gaussian weights of various data types in historical power generation data and historical meteorological data based on the Gaussian weighting function; and then stack the Gaussian weights with the historical power generation data and historical meteorological data along the channel dimension.
[0053] This embodiment can divide wind power prediction into two parts: weather prediction and power prediction. In the weather prediction part, historical weather data is used to predict future weather data, and in the power prediction part, the predicted future weather data is used to predict future wind power generation.
[0054] This embodiment can divide historical meteorological data into multiple time series samples according to time order to construct a dataset for meteorological forecasting.
[0055] This embodiment does not limit the specific method of obtaining time series samples. Generally, each time series sample may include an input sequence and an output sequence. This embodiment does not limit the length of the input and output sequences, and the length of the input and output sequences is generally chosen to be 8. This embodiment also does not limit the time interval of the data in the sequence, and it can generally be 3 hours. That is, in this embodiment, the time length of both the input and output sequences can be 24 hours.
[0056] S102: Train the RMT-LSTM spatiotemporal prediction deep learning network based on the meteorological dataset constructed from time series samples to obtain the trained meteorological numerical spatiotemporal prediction model.
[0057] This embodiment can train the RMT-LSTM spatiotemporal prediction deep learning network based on a meteorological dataset constructed from time series samples to obtain a trained meteorological numerical spatiotemporal prediction model.
[0058] This embodiment does not limit the ratio of training set to test set; generally, the ratio of training set to test set can be 8:2.
[0059] RMT stands for Retentive Networks Meet Vision Transformers. This model is a deep learning model for spatial feature extraction that incorporates Manhattan self-attention and attention decomposition mechanisms. Manhattan self-attention can capture spatial features at different locations through weight adjustment, and can fully extract spatial feature information of long-distance and short-distance dependencies in the data.
[0060] The ST-LSTM structure is a deep learning structure in the spatiotemporal prediction model PredRNN-V2. This structure mainly captures spatiotemporal information in the data by introducing temporal memory state and spatial memory state. At the same time, it adds decoupling loss to enhance the model's ability to capture long-term dependencies. The extraction of spatial feature information is implemented using traditional convolutional neural networks. Convolutional neural networks are usually used to capture short-distance dependent spatial features, but they cannot fully capture long-distance dependent spatial features.
[0061] This embodiment combines the RMT model and the ST-LSTM structure. Specifically, the RMT model can replace the convolutional neural network in the spatial feature information extraction module of the ST-LSTM structure, thereby improving the spatial feature extraction capability of the model and obtaining the RMT-LSTM spatiotemporal prediction deep learning network.
[0062] This embodiment does not limit the specific architecture of the RMT-LSTM spatiotemporal prediction deep learning network; generally, it can be as follows: Figure 2 As shown, the model can be set to have 4 network layers. When the input to the model is historical meteorological data before (including) time t, the model can output predicted meteorological data after time t.
[0063] In this embodiment, the training set in the meteorological dataset can be input into the RMT-LSTM spatiotemporal prediction deep learning network for training, so as to obtain the trained meteorological numerical spatiotemporal prediction model to be tested.
[0064] The test set from the meteorological dataset is input into the meteorological numerical spatiotemporal prediction model for testing. The test results are obtained, and a set threshold is used to determine whether the meteorological numerical spatiotemporal prediction model has been successfully trained. If the test passes, the meteorological numerical spatiotemporal prediction model is obtained; if it fails, it is retrained. The meteorological numerical spatiotemporal prediction model obtained after training is the first sub-model of the wind power generation prediction model.
[0065] In this embodiment, the threshold setting method is not limited. Generally, it can be set so that the average relative prediction error of multiple meteorological data is less than 10%.
[0066] S103: A multi-channel spatial deep learning network is trained on a spatial data conversion dataset constructed based on historical power generation data and historical meteorological data to obtain a spatial data conversion model for power generation.
[0067] This embodiment can train a multi-channel spatial deep learning network based on a spatial data conversion dataset constructed from historical power generation data and historical meteorological data to obtain a spatial data conversion model for power generation.
[0068] This embodiment can construct a sample of a spatial data conversion dataset by collecting historical power generation data and historical meteorological data at the same time. The input data in the sample is historical meteorological data, and the output data in the sample is historical power generation data.
[0069] This embodiment does not limit the ratio of training set to test set in the spatial data conversion dataset; it can generally be 8:2.
[0070] Since wind power generation is most affected by ground weather, this embodiment can train a multi-channel spatial deep learning network based on a spatial data conversion dataset constructed from historical power generation data and historical numerical weather ground data in historical meteorological data, to obtain a spatial data conversion model for power generation.
[0071] In this embodiment, the training set in the spatial data conversion dataset can be input into a multi-channel spatial deep learning network for training to obtain a trained power generation spatial data conversion model. The multi-channel spatial deep learning network can be a CNN (Convolutional Neural Networks) or a Swin-Transformer model. The Swin-Transformer is a new visual Transformer model that can be used as a general backbone for computer vision.
[0072] In this embodiment, the test set from the spatial data conversion dataset can be input into the power generation spatial data conversion model under test for testing. The test results are obtained, and a set threshold is used to determine whether the power generation spatial data conversion model under test has been trained successfully. If the test passes, the power generation spatial data conversion model is obtained; if it fails, it is retrained. The power generation spatial data conversion model obtained after training is the second sub-model of the wind power generation prediction model.
[0073] In this embodiment, the threshold setting method is not limited. Generally, it can be set so that the average relative prediction error of multiple meteorological data is less than 10%.
[0074] S104: Input the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and input the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0075] This embodiment can input the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and input the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0076] That is, the wind power generation model in this embodiment may include two sub-models, namely a meteorological numerical spatiotemporal prediction model and a power generation spatial data conversion model.
[0077] In this embodiment, the meteorological data to be measured in the area where the wind farm is located can be input into the wind power generation prediction model. The data will first pass through the first sub-model of the wind power generation prediction model, namely the meteorological numerical spatiotemporal prediction model, to obtain the prediction results of various meteorological elements in the area where the wind farm is located in the future.
[0078] Furthermore, the surface meteorological data from the prediction results of various meteorological elements can be input into the second sub-model of the wind power generation prediction model, namely the power generation spatial data conversion model, to obtain the prediction results of future wind power generation.
[0079] Based on the above embodiments, the method of the present invention divides the wind power generation prediction problem into two problems: meteorological numerical spatiotemporal prediction and power generation mapping, and solves them sequentially. In the meteorological numerical spatiotemporal prediction process, the RMT-LSTM structure, which integrates the RMT model and the ST-LSTM model, is used to improve the model's ability to extract spatial features and make the prediction results more accurate.
[0080] In this embodiment, the general training process for the meteorological numerical spatiotemporal prediction model can be as follows: Figure 3As shown, this embodiment can obtain raw meteorological data from satellite remote sensing data, numerical weather upper-air data, and numerical weather surface data, and preprocess the raw meteorological data to obtain preprocessed meteorological data. The preprocessing process includes Gaussian weight calculation, constructing a meteorological dataset based on the preprocessed meteorological data, training an RMT-LSTM spatiotemporal prediction deep learning model based on the meteorological dataset, and using the trained RMT-LSTM spatiotemporal prediction deep learning model as the meteorological numerical spatiotemporal prediction model.
[0081] In this embodiment, the general training process for the power generation spatial data conversion model can be as follows: Figure 4 As shown, this embodiment can obtain the original power generation dataset ground meteorological data from the numerical weather ground data of the power generation dataset, preprocess the original power generation dataset ground meteorological data to obtain preprocessed power conversion data, and construct a spatial data conversion dataset based on the preprocessed power conversion data to train a CNN multi-channel spatial deep learning network, and use the trained CNN model as the power generation spatial data conversion model.
[0082] The general process for predicting wind power generation in this embodiment can be as follows: Figure 5 As shown, in this embodiment, meteorological data to be measured can be obtained from real-time collected satellite remote sensing data, numerical weather upper-air data, and numerical weather ground data. The meteorological data to be measured is first input into the meteorological numerical spatiotemporal prediction model to obtain prediction results of various meteorological elements. In this embodiment, the prediction results of ground meteorological data can be selected to predict wind power generation. The prediction results of ground meteorological data are input into the power generation spatial data conversion model to obtain the output wind power generation prediction results.
[0083] The following combination Figure 6 , Figure 6 This is a structural block diagram of a wind power prediction device provided in an embodiment of the present invention. The device may include:
[0084] The first module 100 is used to acquire historical power generation data of the wind farm and historical meteorological data of the target area, and to divide the historical meteorological data into multiple time series samples according to the time sequence.
[0085] The second module 200 is used to train the RMT-LSTM spatiotemporal prediction deep learning network based on the meteorological dataset constructed from the time series samples, so as to obtain the trained meteorological numerical spatiotemporal prediction model.
[0086] The third module 300 is used to train a multi-channel spatial deep learning network based on the spatial data conversion dataset constructed from the historical power generation data and the historical meteorological data to obtain a power generation spatial data conversion model.
[0087] The fourth module 400 is used to input the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and to input the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results.
[0088] Based on the above embodiments, the method of the present invention divides the wind power generation prediction problem into two problems: meteorological numerical spatiotemporal prediction and power generation mapping, and solves them sequentially. In the meteorological numerical spatiotemporal prediction process, the RMT-LSTM structure, which integrates the RMT model and the ST-LSTM model, is used to improve the model's ability to extract spatial features and make the prediction results more accurate.
[0089] Based on the above embodiments, the first module 100 may include:
[0090] The first unit is used to acquire the initial historical power generation data of the wind farm and the initial historical meteorological data of the target area;
[0091] The second unit is used to perform interpolation processing on the missing data in the initial historical power generation data and the initial historical meteorological data to obtain complete historical power generation data and complete historical meteorological data.
[0092] The third unit is used to standardize the supplemented historical power generation data and the supplemented historical meteorological data to obtain the historical power generation data and the historical meteorological data.
[0093] Based on the above embodiments, the first module 100 may further include:
[0094] The fourth unit is used to determine the Gaussian weights of each type of data in the historical meteorological data based on the Gaussian weighting function.
[0095] The fifth unit is used to stack the Gaussian weights and the historical meteorological data in the channel dimension.
[0096] Based on the above embodiments, the historical meteorological data includes: historical satellite remote sensing data, historical numerical weather upper-air data, and historical numerical weather surface data.
[0097] Based on the above embodiments, the third module 300 includes:
[0098] The sixth unit is used to train the multi-channel spatial deep learning network based on the spatial data conversion dataset constructed from the historical power generation data and the historical numerical weather surface data in the historical meteorological data, so as to obtain the power generation spatial data conversion model.
[0099] Based on the above embodiments, the RMT-LSTM spatiotemporal prediction deep learning network has 4 layers.
[0100] Based on the above embodiments, the multi-channel spatial deep learning network is a CNN model or a Swin-Transformer model.
[0101] Based on the above embodiments, the present invention also provides an electronic device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the device may also include various necessary network interfaces, a power supply, and other components.
[0102] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by an execution terminal or processor, can implement the method provided in the embodiments of the present invention; the storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0103] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
Claims
1. A method for predicting wind power output, characterized in that, include: Acquire historical power generation data of the wind farm and historical meteorological data of the target area, and divide the historical meteorological data into multiple time series samples according to time sequence; The RMT-LSTM spatiotemporal prediction deep learning network is trained on the meteorological dataset constructed based on the time series samples to obtain a trained meteorological numerical spatiotemporal prediction model; wherein, the RMT-LSTM is a network obtained by replacing the convolutional neural network in the ST-LSTM structure spatial feature information extraction module with the RMT model. A multi-channel spatial deep learning network is trained on a spatial data conversion dataset constructed based on the historical power generation data and the historical meteorological data to obtain a spatial data conversion model for power generation. The meteorological data to be measured is input into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and the meteorological element prediction results are input into the power generation spatial data conversion model to obtain the output power generation prediction results. The acquisition of historical power generation data of the wind farm and historical meteorological data of the target area includes: Obtain the initial historical power generation data of the wind farm and the initial historical meteorological data of the target area; Interpolation processing is performed on the missing data in the initial historical power generation data and the initial historical meteorological data to obtain complete historical power generation data and complete historical meteorological data. The historical power generation data and the historical meteorological data are obtained by standardizing the complete historical power generation data and the historical meteorological data. The Gaussian weights of each type of data in the historical meteorological data are determined based on the Gaussian weighting function. The Gaussian weights are stacked with the historical meteorological data in the channel dimension.
2. The wind power prediction method according to claim 1, characterized in that, The historical meteorological data includes: historical satellite remote sensing data, historical numerical weather upper-air data, and historical numerical weather surface data.
3. The wind power prediction method according to claim 1, characterized in that, The spatial data conversion dataset constructed based on the historical power generation data and the historical meteorological data is used to train a multi-channel spatial deep learning network to obtain a spatial data conversion model for power generation, including: The multi-channel spatial deep learning network is trained using the spatial data conversion dataset constructed based on the historical power generation data and the historical numerical weather surface data in the historical meteorological data to obtain the spatial data conversion model for power generation.
4. The wind power prediction method according to claim 1, characterized in that, The RMT-LSTM spatiotemporal prediction deep learning network has 4 layers.
5. The wind power prediction method according to claim 1, characterized in that, The multi-channel spatial deep learning network is a CNN model or a Swing-Transformer model.
6. A wind power prediction device, characterized in that, include: The first module is used to acquire historical power generation data of the wind farm and historical meteorological data of the target area, and to divide the historical meteorological data into multiple time series samples according to the time sequence. The second module is used to train the RMT-LSTM spatiotemporal prediction deep learning network based on the meteorological dataset constructed from the time series samples, so as to obtain the trained meteorological numerical spatiotemporal prediction model; wherein, the RMT-LSTM is a network obtained by replacing the convolutional neural network in the ST-LSTM structure spatial feature information extraction module with the RMT model. The third module is used to train a multi-channel spatial deep learning network based on the spatial data conversion dataset constructed from the historical power generation data and the historical meteorological data, so as to obtain the power generation spatial data conversion model. The fourth module is used to input the meteorological data to be measured into the meteorological numerical spatiotemporal prediction model to obtain the output meteorological element prediction results, and to input the meteorological element prediction results into the power generation spatial data conversion model to obtain the output power generation prediction results. The acquisition of historical power generation data of the wind farm and historical meteorological data of the target area includes: Obtain the initial historical power generation data of the wind farm and the initial historical meteorological data of the target area; Interpolation processing is performed on the missing data in the initial historical power generation data and the initial historical meteorological data to obtain complete historical power generation data and complete historical meteorological data. The historical power generation data and the historical meteorological data are obtained by standardizing the complete historical power generation data and the historical meteorological data. The Gaussian weights of each type of data in the historical meteorological data are determined based on the Gaussian weighting function. The Gaussian weights are stacked with the historical meteorological data in the channel dimension.
7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the wind power prediction method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the wind power prediction method as described in any one of claims 1 to 5.
Citation Information
Patent Citations
Wind power generation power prediction method and device
CN118114809A
New energy wind-solar power prediction method and device based on large meteorological model
CN119204349A
