Construction method, model and system of downscaling wind power plant wind speed prediction model
By constructing a downscale wind speed prediction model for wind power farms, using mesoscale weather forecast mode and mode climate data, the problem of difficult to provide fine wind energy research information in local wind farms in the existing technology is solved, and the effect of fine wind speed prediction and reducing the risk of wind power development is achieved.
Patent Information
- Application Number
- CN202411474328.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-25
AI Technical Summary
The existing technology is difficult to provide local wind energy research information for local wind farms, which has led to wind power development facing the challenges of uncertainty in wind energy resources brought about by climate change.
A downscale wind speed prediction model is constructed, and the wind parameter simulation data is obtained using the mesoscale weather forecast mode, and the representative annual correction is performed. Sample data sets are constructed based on the pattern climate data, and a high-efficiency sub-pixel convolutional neural network is used for training to generate a downscale wind speed prediction model.
It has achieved the provision of fine wind speed prediction information for local wind farms, reduced the risks of wind power development, and supported wind farm site selection and long-term planning.
Smart Images
Figure CN119514322A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of atmospheric science and technology, and in particular to a method for constructing a downscaled wind farm wind speed prediction model, a model and a system. Background Art
[0002] With the development and utilization of wind energy resources and the large-scale construction of wind farms in China, the long-term operation of wind farms faces the challenge of uncertainty in wind energy resources caused by climate change. Previous studies have shown that surface wind speeds in the Northern Hemisphere, including China, have declined in the past few decades, and global / regional climate models (GCMs / RCMs) have been widely used to predict climate change related to surface wind speed. Based on the output data of these models, especially from CMIP6 (the sixth phase of the International Coupled Model Intercomparison Project), it is possible to study the evolution of wind energy in wind farms. The spatial resolution of CMIP6 (about 100 kilometers) is too coarse to provide detailed information for local wind energy research. Summary of the invention
[0003] In view of the defects in the prior art, the present invention provides a method, model and system for constructing a downscaled wind farm wind speed prediction model, so as to provide detailed information for local wind energy research and reduce the risk of wind power development.
[0004] A method for constructing a downscaled wind farm wind speed prediction model, comprising:
[0005] The mesoscale weather forecast model is used to obtain wind parameter simulation data at the hub height of wind turbines in the target wind farm area for at least one complete historical year, and a simulated wind speed set is obtained;
[0006] Based on the preset data set at the hub height of the wind turbine in the target wind farm area, the simulated wind speed in the simulated wind speed set is corrected for the representative year to obtain a background wind spectrum data set;
[0007] Sample data is constructed according to the model climate data and the wind speed prediction results of the target wind farm to obtain a sample data set;
[0008] The background wind spectrum data in the background wind spectrum data set is used to construct the loss function to obtain the target loss function;
[0009] The sample data in the sample data set is used to train the efficient sub-pixel convolutional neural network until the target loss function converges, and then the downscaled wind farm wind speed prediction model is obtained.
[0010] Furthermore, it also includes normalizing the background wind spectrogram data in the background wind spectrogram data set.
[0011] Furthermore, the efficient sub-pixel convolutional neural network includes four convolutional layers and one sub-pixel convolutional layer.
[0012] Furthermore, the convolutional layer of the efficient sub-pixel convolutional neural network is represented by the following formula:
[0013]
[0014] Among them, X and Y are the input and output of the convolutional layer of the efficient sub-pixel convolutional neural network, W and B are the weight and bias, respectively. is a nonlinear function.
[0015] Furthermore, the model climate data includes primary model climate data and auxiliary model climate data, wherein:
[0016] The main model climate data include temperature data, relative humidity data and wind speed field data from ERA5 or CMIP6;
[0017] Auxiliary model climate data include coarse-resolution and fine-resolution terrain data, and coarse-resolution and fine-resolution surface roughness data.
[0018] Furthermore, the topographic data include the mean altitude and the standard deviation of the altitude.
[0019] Furthermore, the cost function of the downscaled wind farm wind speed prediction model is the mean square error.
[0020] A downscaled wind farm wind speed prediction model is constructed by using any of the above-mentioned methods for constructing a downscaled wind farm wind speed prediction model.
[0021] A downscaled wind farm wind speed prediction and analysis system uses a downscaled wind farm wind speed prediction model to predict the wind speed of a target wind farm; it includes an acquisition module, a correction module and a prediction module, wherein:
[0022] The acquisition module is used to obtain wind parameter simulation data at the hub height of the wind turbine in the target wind farm area for at least one complete historical year using a mesoscale weather forecast model to obtain a simulated wind speed set;
[0023] The correction module is used to perform representative year correction processing on the simulated wind speed in the simulated wind speed set based on a preset data set at the hub height of the wind turbine in the target wind farm area to obtain a background wind spectrum data set;
[0024] The prediction module is used to predict the wind speed of the target wind farm using the downscaled wind farm wind speed prediction model according to the model climate data. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] In order to more clearly illustrate the specific embodiments of the present invention, the following will briefly introduce the drawings required for the specific embodiments or the prior art description. In all the drawings, each element or part is not necessarily drawn according to the actual scale.
[0026] Figure 1 A flowchart of a method for constructing a downscaled wind farm wind speed prediction model provided by an embodiment of the present invention;
[0027] Figure 2 A structural diagram of a downscaled wind farm wind speed prediction model provided by an embodiment of the present invention;
[0028] Figure 3 A schematic diagram of error analysis between a measured wind speed result of a target wind farm and a predicted wind speed result obtained by predicting the target wind farm using a downscaled wind farm wind speed prediction model provided in an embodiment of the present invention;
[0029] Figure 4 A schematic diagram of the spatial distribution of a simulated wind speed result of a target wind farm and a wind speed prediction result obtained by predicting the target wind farm using a downscaled wind farm wind speed prediction model provided in an embodiment of the present invention;
[0030] Figure 5 The monthly average wind speed change of the wind speed prediction result obtained by processing the historical monthly average wind speed (1978-2014) and the model climate data of CMIP6 in the medium-term (2015-2050) in the target wind farm area provided by an embodiment of the present invention with the downscaled wind farm wind speed prediction model is provided. DETAILED DESCRIPTION
[0031] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.
[0032] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.
[0033] In one embodiment, if Figure 1 As shown, a method for constructing a downscaled wind farm wind speed prediction model is provided, including:
[0034] 1. Use the mesoscale weather forecast model to obtain wind parameter simulation data at the hub height of the wind turbine in the target wind farm area for at least one complete historical year to obtain a simulated wind speed set;
[0035] The mesoscale weather forecast model (WRF) is used to obtain the wind parameter simulation data at the hub height of the wind turbine in the target wind farm area for at least one complete historical year, and the simulated wind speed set is obtained.
[0036] The wind parameter simulation data at the height of the wind turbine hub is the data of wind parameters such as wind speed and wind direction obtained by simulation of mathematical models or computer programs. Each simulated data point in the simulated wind speed set has a fine spatial resolution covering 1 km or 0.01° on the geographic coordinates. Such fine spatial resolution can capture smaller geographic features or changes or provide relatively detailed wind speed and wind direction information within the target wind farm area.
[0037] The Mesoscale Weather Forecast Model (WRF) is an advanced numerical weather forecast model that can simulate and predict various mesoscale weather phenomena, such as heavy rain, tornadoes, strong winds, etc. It has high spatial and temporal resolution and can provide more detailed weather forecast information.
[0038] The wind turbine hub height refers to the height of the wind turbine hub from the ground. The wind turbine hub is one of the key components of the wind turbine. It connects the blades and the wind turbine, and plays a role in supporting and transmitting torque.
[0039] 2. Based on the preset data set at the hub height of the wind turbine in the target wind farm area, the simulated wind speed in the simulated wind speed set is corrected for the representative year to obtain the background wind spectrum data set;
[0040] Because the simulated wind speed results of WRF are all historical data, which are historical data of at least one full year, the simulated wind speed set may have different wind years and cannot fully reflect the annual changes of actual wind speed. If this simulated wind speed set is used as the training target of the downscaled wind farm wind speed prediction model, it will lead to great deviations in the trained model.
[0041] Therefore, it is necessary to use a preset data set of more than 20 years to correct the simulated wind speed for representative years (i.e. years with typical wind speed characteristics) to eliminate the impact of climate and obtain a background wind spectrum data set, so that the trained downscaled wind farm wind speed prediction model can be more accurate and reasonable.
[0042] The preset dataset is ERA5 mesoscale long-term series wind speed data, where ERA5 is a global climate and weather reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), which contains high-resolution data of multiple meteorological variables, including wind speed, wind direction, temperature, etc. Mesoscale refers to the spatial scale of weather phenomena described in meteorology, usually between the weather scale (such as fronts, cyclones) and the microscale (such as valley winds, urban heat islands). Long-term series wind speed data refers to wind speed data collected continuously over a period of time (which may be several years or longer).
[0043] Preferably, the method further includes normalizing the background wind spectrum data in the background wind spectrum data set.
[0044] 3. Construct sample data based on model climate data and wind speed prediction results of the target wind farm to obtain a sample data set;
[0045] Preferably, the model climate data includes primary model climate data and auxiliary model climate data, wherein:
[0046] The main model climate data include temperature data, relative humidity data and wind speed field data from ERA5 or CMIP6;
[0047] CMIP6 refers to the sixth phase of the International Coupled Model Intercomparison Project.
[0048] Auxiliary model climate data include coarse-resolution and fine-resolution terrain data, and coarse-resolution and fine-resolution surface roughness data.
[0049] Preferably, the terrain data further includes the average altitude and the altitude standard deviation.
[0050] 4. Use the background wind spectrum data in the background wind spectrum data set to construct the loss function and obtain the target loss function;
[0051] The background wind spectrum data in the background wind spectrum dataset is used as the training target of the downscaled wind farm wind speed prediction model. The wind spectrum is also called the wind speed map. The wind spectrum shows the frequency distribution characteristics of wind speed. It is usually drawn with wind speed as the vertical coordinate and frequency (or period) as the horizontal coordinate. This kind of graph can help understand the energy distribution of wind farms at different frequencies, so as to further analyze the characteristics and potential of wind resources.
[0052] 5. The sample data in the sample data set is used to train the efficient sub-pixel convolutional neural network until the target loss function converges, and the downscaled wind farm wind speed prediction model is obtained.
[0053] The efficient sub-pixel convolutional neural network is a network architecture rebuilt on the basis of ESPCN (Efficient Sub-Pixel Convolutional Neural Network).
[0054] The model climate data is used as the input of the efficient sub-pixel convolutional neural network, the wind speed prediction result of the target wind farm is used as the output, and the background wind spectrum data in the background wind spectrum data set is used as the training target. The efficient sub-pixel convolutional neural network is trained to obtain the downscaled wind farm wind speed prediction model.
[0055] Preferably, the efficient sub-pixel convolutional neural network includes four convolutional layers and one sub-pixel convolutional layer, and the first two convolutional layers are used to extract coarse-resolution features to obtain coarse-resolution features, that is, features extracted at coarse resolution.
[0056] Sub-pixel convolutional layers are used to reconstruct the extracted coarse-resolution features to generate initial high-resolution features.
[0057] The third convolutional layer is used to further extract the initial high-resolution features to obtain high-resolution features.
[0058] The fourth convolutional layer is an efficient sub-pixel convolutional neural network linear mapping layer, which is used to linearly map the high-resolution features to generate a high-resolution wind spectrum.
[0059] Each of the four convolutional layers of the efficient sub-pixel convolutional neural network can be expressed by the following formula, but the sizes of X, Y, W, and B are as follows: Figure 2 Shown is different.
[0060]
[0061] Among them, X and Y are the input and output of the convolutional layer, respectively, and W and B are the weights (i.e. filters) and biases that can be learned through training, respectively. Is a nonlinear function, also called an activation function.
[0062] Specifically, Figure 2 a represents the structure of efficient sub-pixel convolutional neural network, Figure 2 b is a schematic diagram showing the convolution of the convolutional layer of the efficient sub-pixel convolutional neural network. Figure 2 c Schematic diagram showing convolutions performed by the sub-pixel convolutional layer of the efficient sub-pixel convolutional neural network.
[0063] Figure 2 Cov2D in a represents convolution, and SubpixelCov2D represents subpixel convolution. The filter size in each convolution is given in the brackets after "Cov2D". Ele_std_c and Ele_std_f represent the elevation maps of coarse resolution and fine resolution, respectively; Ele_std_c and Ele_std_f represent the standard deviation maps of elevation within each coarse grid and each fine grid, respectively; P_c is the coarse resolution wind speed map; P_f is the high-resolution wind speed map (high-resolution wind spectrum map); ns is the sample size (the monthly wind speed map is regarded as a sample); nx and ny represent the dimensions of the input coarse resolution wind speed map (model climate data) in the x direction and the y direction, respectively; R is the ratio of resolution enhancement, that is, the ratio of coarse resolution to fine resolution.
[0064] Preferably, in an embodiment provided in the present invention, a ReLU activation function is used, which returns the maximum value between 0 and W*X+B.
[0065] Resolution enhancement of model climate data is achieved through a sub-pixel convolutional layer, in which an R 2The feature maps are reshaped into a single map of size (nx×R)×(ny×R), as Figure 2 c. Here R is the ratio between coarse resolution and fine resolution.
[0066] The cost function for training the downscaled wind farm wind speed prediction model is the mean square error. Different mean square error values are used to stop training: first, the model is trained for 1000 epochs (traversal, one epoch is equal to inputting all samples in the sample data set into the efficient sub-pixel convolutional neural network for a complete training cycle) to ensure that the mean square error reaches a stable level; second, the mean square error of the last epoch should be less than one tenth of the mean square error of the first epoch to prevent the model from falling into a local minimum. Once these criteria are met, the training of the downscaled wind farm wind speed prediction model is stopped, otherwise, the model will be trained for another 1000 epochs.
[0067] In addition, in order to reduce the number of parameters that need to be optimized during the training of the downscaled wind farm wind speed prediction model, the training of the model is divided into two steps, namely, first downscaling the model climate data with a spatial resolution of 0.25° to 0.1° (spatial resolution of 0.1°), and then downscaling the model climate data with a spatial resolution of 0.1° to 0.01° (spatial resolution of 0.01°).
[0068] After the model training is completed, the main model climate data such as surface wind speed data of ERA5 and CMIP6 (including temperature data, relative humidity data and wind speed field data) can be input into the trained downscaled wind farm wind speed prediction model to generate high-resolution wind spectrum data. Thus, the coarse-resolution model climate data is downscaled to monthly and kilometer-level surface wind speed data to generate high-resolution wind spectrum data.
[0069] The downscaled wind farm wind speed prediction model is built using Tensorflow (an open source machine learning framework) and Tensorlayer (TensorLayer is an advanced encapsulation library of TensorFlow for building custom layers).
[0070] like Figure 3 As shown, the wind speed prediction method based on the downscaled wind farm wind speed prediction model proposed in the present invention has an absolute deviation of less than 4% between the wind speed prediction result of the target wind farm and the measured wind speed result.
[0071] The present invention has been tested on multiple wind towers in the target wind farm. Table 1 shows the error analysis of the wind speed prediction results of the present invention on the four wind towers and the simulated wind speed results of WRF.
[0072] Table 1 is the error statistics for verifying the 100-meter predicted wind speed results obtained by the downscaled wind farm wind speed prediction model based on the model climate data of CMIP6 (ssp245 and ssp585) based on the measured wind speed results of the wind towers (1822#, 1823#, 2137# and 2138#) and the simulated wind speed results obtained by WRF numerical simulation (reference wind speed data).
[0073] Table 1
[0074]
[0075] Figure 4 The spatial distribution of wind speed prediction results obtained by processing the WRF simulated wind speed results and the CMIP6 model climate data through the downscaled wind farm wind speed prediction model is shown.
[0076] Figure 5 The monthly average wind speed changes in the target wind farm area are obtained by processing the historical monthly average wind speed (1978-2014) and the future medium-term (2015-2050) CMIP6 model climate data with the downscaled wind farm wind speed prediction model.
[0077] The present invention utilizes the strong correlation between simulated wind speed and model climate data obtained by numerical simulation, applies the efficient feature extraction and data mining capabilities of deep learning in big data processing, and uses the downscaled wind farm wind speed prediction model to carry out wind speed prediction for a target wind farm at a specific site. The temporal and spatial variation characteristics of future wind speed at a specific site are analyzed, providing a scientific basis for site selection and long-term planning of wind farms, reducing the risks of wind power development, and supporting reasonable planning of decision makers.
[0078] In one embodiment, a downscaled wind farm wind speed prediction model is provided, which is constructed by using any of the above-mentioned methods for constructing a downscaled wind farm wind speed prediction model.
[0079] In one embodiment, a downscaled wind farm wind speed prediction and analysis system is provided, which uses a downscaled wind farm wind speed prediction model to predict the wind speed of a target wind farm; it includes an acquisition module, a correction module and a prediction module, wherein:
[0080] The acquisition module is used to obtain wind parameter simulation data at the hub height of the wind turbine in the target wind farm area for at least one complete historical year using a mesoscale weather forecast model to obtain a simulated wind speed set;
[0081] The correction module is used to perform representative year correction processing on the simulated wind speed in the simulated wind speed set based on a preset data set at the hub height of the wind turbine in the target wind farm area to obtain a background wind spectrum data set;
[0082] The prediction module is used to predict the wind speed of the target wind farm using the downscaled wind farm wind speed prediction model according to the model climate data.
[0083] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. 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 replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.
Claims
1. A method for constructing a downscaled wind farm wind speed prediction model, characterized in that: include: The mesoscale weather forecast model is used to obtain wind parameter simulation data at the hub height of wind turbines in the target wind farm area for at least one complete historical year, and a simulated wind speed set is obtained; Based on the preset data set at the wind turbine hub height in the target wind farm area, the simulated wind speed in the simulated wind speed set is corrected for a representative year to obtain a background wind spectrum data set; Sample data is constructed according to the model climate data and the wind speed prediction results of the target wind farm to obtain a sample data set; The background wind spectrum data in the background wind spectrum data set is used to construct the loss function to obtain the target loss function; The sample data in the sample data set is used to train the efficient sub-pixel convolutional neural network until the target loss function converges, thereby obtaining a downscaled wind farm wind speed prediction model.
2. A method for downscaling a wind farm wind speed prediction model according to claim 1, characterized in that: It also includes normalizing the background wind spectrogram data in the background wind spectrogram data set.
3. A method for downscaling a wind farm wind speed prediction model according to claim 1 or 2, characterized in that: The efficient sub-pixel convolutional neural network includes four convolutional layers and one sub-pixel convolutional layer.
4. A method for downscaling a wind farm wind speed prediction model according to claim 3, characterized in that: Also includes: The convolutional layer of the efficient sub-pixel convolutional neural network is represented by the following formula: Among them, X and Y are the input and output of the convolutional layer of the efficient sub-pixel convolutional neural network, W and B are the weight and bias, respectively. is a nonlinear function.
5. A method for downscaling a wind farm wind speed prediction model according to claim 4, characterized in that: The model climate data includes primary model climate data and auxiliary model climate data, wherein: The main model climate data include temperature data, relative humidity data and wind speed field data of ERA5 or CMIP6; The auxiliary model climate data includes terrain data with coarse resolution and fine resolution, and surface roughness data with coarse resolution and fine resolution.
6. A method for downscaling a wind farm wind speed prediction model according to claim 5, characterized in that: The terrain data includes the average altitude and the standard deviation of the altitude.
7. A method for downscaling a wind farm wind speed prediction model according to claim 6, characterized in that: The cost function of the downscaled wind farm wind speed prediction model is the mean square error.
8. A downscaled wind farm wind speed prediction model, characterized in that: The method is constructed by the construction method described in claims 1 to 7.
9. A downscaled wind farm wind speed prediction and analysis system, characterized in that: The downscaled wind farm wind speed prediction model of claim 8 is used to predict the wind speed of the target wind farm; the downscaled wind farm wind speed prediction model comprises an acquisition module, a correction module and a prediction module, wherein: The acquisition module is used to obtain wind parameter simulation data at the hub height of the wind turbine in the target wind farm area for at least one complete historical year using a mesoscale weather forecast model to obtain a simulated wind speed set; The correction module is used to perform representative year correction processing on the simulated wind speed in the simulated wind speed set based on the preset data set at the hub height of the wind turbine in the target wind farm area to obtain a background wind spectrum data set; The prediction module is used to predict the wind speed of the target wind farm using a downscaled wind farm wind speed prediction model according to the model climate data.