Local surface layer wind field fusion and nowcasting method
By combining the wind shear index of the meteorological tower with Barnes stepwise correction and a convolutional neural network model, the problem of integrating WRF-LES large eddy simulation with actual wind fields was solved, achieving efficient forecasting of local near-surface wind fields, which is applicable to wind field forecasting for wind power and low-altitude economics.
Patent Information
- Application Number
- CN202511573298.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-30
- Publication Date
- 2026-02-03
AI Technical Summary
In existing technologies, there are few studies on the integration of WRF-LES large eddy simulation with actual wind fields, resulting in insufficient local near-surface wind field forecasts. In particular, the utilization rate of wind field data in fields such as wind power and low-altitude flight is not high, and the application of convolutional neural networks in wind field and large eddy forecasting is limited.
By using the wind shear index calculated by the wind measurement tower to fill the gaps in the wind profile above the automatic weather station, and combining the WRF-LES large eddy simulation and automatic weather station data to perform Barnes stepwise correction wind field fusion, a convolutional neural network model is established to perform wind field data fusion and nowcasting.
It achieves efficient fusion and nowcasting of local near-surface wind fields, improving the accuracy of wind field forecasts in fields such as wind power and low-altitude economy, and uses model parameters in WRF-LES real-time forecasts for correction.
Smart Images
Figure CN121456348A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of meteorological wind field fusion and forecasting technology, specifically involving a local near-surface wind field fusion and nowcasting method. Background Technology
[0002] With economic and technological development, the forecasting of meteorological conditions is becoming increasingly important. In recent years, fields such as wind power and low-altitude flight have placed higher demands on the forecasting of low-altitude wind fields. Generally speaking, meteorological equipment such as wind measurement towers and automatic weather stations are built in advance for wind farm construction and other related fields. However, in actual wind energy assessment and forecasting, the data from the wind measurement towers are mainly referred to, while the data from the automatic weather stations around the wind measurement towers are not utilized sufficiently.
[0003] Current forecasts of near-surface wind fields mainly focus on wind power and some specialized industries. The methods employed are primarily objective correction and deep learning. Because WRF-LES large eddy simulation forecasts have high computational requirements, necessitating high-resolution meteorological models as input, they are rarely used in operational forecasting, being primarily employed for research and operational review simulations. Research on the integration of WRF-LES large eddy simulations with actual wind fields is scarce, and studies on forecast correction based on this integration are also limited.
[0004] Data fusion of wind measurement towers, automatic weather station data, and model wind fields is relatively rare, mainly due to the limited number of stations and insufficient model resolution. If there are enough stations and the model's calculation grid is sufficiently fine, the fusion of local wind fields is achievable.
[0005] With the increasing prevalence of high-resolution numerical weather prediction products, WRF-LES large eddy forecasting will become possible. Furthermore, while convolutional neural networks are widely used in weather forecasting, their application in modeling fused wind fields and large eddy forecast wind fields is still limited. Therefore, it is necessary to integrate and fuse wind fields and large eddy simulation wind fields, conduct convolutional network modeling, and use the model parameters as the basis for near-term wind field forecasting. Summary of the Invention
[0006] (a) Technical problems to be solved
[0007] The technical problem to be solved by this invention is how to provide a method for local near-surface wind field fusion and nowcasting, so as to solve the problems of fusion of WRF-LES large eddy simulation and actual wind field and nowcasting wind field.
[0008] (II) Technical Solution
[0009] To address the aforementioned technical problems, this invention proposes a method for local near-surface wind field fusion and nowcasting, which includes the following steps:
[0010] S1. Wind field gap filling above automatic weather stations: The wind shear index calculated by the wind measurement tower is used to fill the gap in the wind profile above the surrounding automatic weather stations.
[0011] S2, Wind Field Fusion: Combine the forecast product dataset of WRF-LES large eddy simulation with the wind fields above tower wind measurement and automatic weather station sites to carry out Barnes stepwise correction wind field fusion to obtain the fused product dataset;
[0012] S3. Convolutional Neural Network Modeling and Near-Term Correction Forecast: A convolutional neural network model is established and trained between the forecast product dataset and the fused product dataset. In actual use, the WRF-LES forecast product is input into the trained convolutional neural network model to obtain the corrected forecast wind field.
[0013] (III) Beneficial Effects
[0014] This invention proposes a method for local near-surface wind field fusion and nowcasting. The invention obtains the local wind field profile by using the power law formula of near-surface wind field data from wind towers and automatic weather stations. Then, the wind field is fused using WRF-LES. Based on this, convolutional neural network modeling is carried out to obtain model parameters, which are then used in WRF-LES real-time forecasting to obtain nowcasting corrections of the near-surface wind field.
[0015] This invention makes full use of local wind measurement tower and encrypted automatic weather station data, and integrates WRF-LES modeling, which can be used for local wind field forecasting. The corresponding algorithms and processes are applicable to wind field forecasting for wind power, low-altitude economy, etc. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the local near-surface wind field fusion and nowcasting method of the present invention. Detailed Implementation
[0017] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
[0018] This invention belongs to the field of local meteorological wind field fusion and forecasting technology. Specifically, it includes the fusion of wind field data from wind towers, automatic weather stations and WRF-LES large eddy simulation using Barnes stepwise correction technology, and a near-surface wind field nowcasting method that integrates convolutional neural networks and fused wind fields.
[0019] This invention proposes using the wind shear index calculated by a 90-meter anemometer tower to fill gaps in the wind profile at a height of 90 meters above surrounding automatic weather stations. It combines this with WRF-LES large eddy simulation to perform Barnes-corrected wind field fusion, and establishes a convolutional neural network model between the large eddy simulation dataset and the fused wind field dataset. In practical applications, the WRF-LES forecast combined with the convolutional model yields corrected wind field forecasts for nearby locations.
[0020] This invention proposes a method for local near-surface wind field fusion and nowcasting, which specifically includes the following steps:
[0021] S1. Wind field gap filling above automatic weather stations: The wind shear index calculated by the wind measurement tower is used to fill the gap in the wind profile above the surrounding automatic weather stations.
[0022] S2, Wind Field Fusion: Combine the forecast product dataset of WRF-LES large eddy simulation with the wind fields above tower wind measurement and automatic weather station sites to carry out Barnes stepwise correction wind field fusion to obtain the fused product dataset;
[0023] S3. Convolutional Neural Network Modeling and Near-Term Correction Forecast: A convolutional neural network model is established and trained between the forecast product dataset and the fused product dataset. In actual use, the WRF-LES forecast product is input into the trained convolutional neural network model to obtain the corrected forecast wind field.
[0024] Example 1:
[0025] S1. Wind field gap filling above automatic weather stations: The wind shear index calculated by the 90-meter anemometer tower is used to fill the gap in the wind profile at a height of 90 meters above the surrounding automatic weather stations.
[0026] S11. Data Quality Control: Measure and acquire meteorological elements at the wind measurement height of each wind tower, remove wind field data during the period directly affected by the typhoon, and remove wind field data with wind speeds deviating excessively from the average wind speed.
[0027] The wind measurement tower has six layers at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters. Each layer contains wind sensors, temperature sensors, and humidity sensors to measure meteorological elements such as wind speed, wind direction, temperature, and relative humidity. Meteorological data is obtained every 10 minutes. Based on the China Meteorological Administration's optimal tropical cyclone path dataset (website: tcdata.typhoon.org.cn), wind field data during the period directly affected by the typhoon has been removed. Using 50 meters as the standard layer, wind speed at 50 meters is measured. Calculate the mean of the sequence Standard deviation Further increase wind speed The wind field value was removed, and the corresponding data at the 10-meter, 20-meter, 35-meter, 70-meter, and 90-meter levels at the same time were also removed.
[0028] S12. Calculation of wind shear index at each level of the anemometer tower: Based on the vertical temperature gradient, the wind field is categorized into three classes—unstable, neutral, and stable—according to atmospheric stability. Using the power-law formula for the vertical distribution of near-surface winds, the wind shear indices for u-wind and v-wind under different stability conditions are calculated. u-wind represents east-west wind, and v-wind represents north-south wind.
[0029] Select the temperatures T at the 20-meter and 70-meter levels of the wind measurement tower and calculate the vertical temperature gradient. ,according to For values less than 0, equal to 0, or greater than 0, atmospheric stability is classified into three categories: unstable, neutral, and stable. Based on changes in atmospheric stability, the wind field at the corresponding moment is included in the samples of the unstable, neutral, and stable categories.
[0030] The wind field is horizontally decomposed into u-wind (east-west wind) and v-wind (north-south wind). For unstable, neutral, and stable samples, based on the power-law formulas for the vertical distribution of near-surface wind (Formulas 1 and 2), a reference height of 10 meters is taken. The corresponding known reference winds are respectively and Calculate the wind shear index of winds at altitudes u and v of 20m, 35m, 50m, 70m, and 90m under different atmospheric stability conditions. ,in and Let z represent the u and v winds at altitudes of 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters, respectively.
[0031] (1)
[0032] (2)
[0033] Calculate the wind shear index at different altitudes The average value is used as the wind shear index for different atmospheric stability at different altitudes (u and v).
[0034] S13. Filling in the gaps in the wind profile from above the automatic weather station to a height of 90 meters: Centered on the anemometer tower, collect and organize real-time wind field observation data from automatic weather stations around the tower. Based on the atmospheric stability and wind shear index of the anemometer tower, calculate the U-wind and V-wind from above the automatic weather station to a height of 90 meters.
[0035] Centered on the wind measuring tower, a square distribution with a radius of 4 kilometers was established to collect and organize real-time wind field observation data from 10 automatic weather stations located in the four directions of east, south, west, and north of the tower. Based on the standards for automatic weather station construction, all measured wind fields were 10 meters high, and the wind fields were decomposed into U-wind and V-wind.
[0036] Atmospheric stability is calculated based on the temperatures at 20 meters and 70 meters of the anemometer tower at the same time. Based on the unstable, neutral, and stable atmospheric stability and wind shear index, the power law formulas of equations (1) and (2) are used to calculate the u-wind and v-wind at heights of 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters above each automatic weather station within 4 kilometers of the anemometer tower.
[0037] S2. Wind Field Fusion: Combining WRF-LES large eddy simulation with tower anemometers and wind fields above automatic weather stations to carry out Barnes-corrected wind field fusion.
[0038] S21. WRF-LES Design and Forecast: Using the 1KM output data of the WRF mode as the input data of the WRF-LES module, the parameters of the WRF-LES module are designed, and the corresponding forecast products are collected and organized into a forecast product dataset.
[0039] The WRF (Weather Research and Forecasting Model) is primarily used for weather-scale simulation and forecasting, while LES (Large Eddy Simulation) is used for large eddy simulation. WRF-LES is a module within the WRF model used for large eddy simulation and forecasting.
[0040] The 1km output data of the WRF model was used as the input data for the WRF-LES module. In the WRF-LES module design: the height distribution below 90 meters was set to 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters; the model's horizontal region was triple-nested, with the grid resolution ratio from the outermost to the innermost layer set to 25:5:1, and the innermost grid resolution set to 40 meters, essentially coinciding with a 4km square area centered on the meteorological tower; the model's forecast times were 05:00, 11:00, 17:00, and 23:00; the integration time was 3 hours, with results output every hour. The U and V wind forecast products for the third hour were then collected, corresponding to 08:00, 14:00, 20:00, and 02:00 the following day, respectively. The corresponding forecast products were collected and organized into a forecast product dataset.
[0041] S22. Fusion of wind fields from WRF-LES forecasts with tower anemometers and automatic weather station data: The Barnes stepwise correction method is adopted. The WRF-LES module calculation results are used as the background field, and the wind fields supplemented by tower anemometers and automatic weather station data are used as the observation field. The fused wind field is obtained after correction, and the fused product dataset is obtained.
[0042] The Barnes stepwise correction method was used to achieve wind field fusion. The calculation results of the WRF-LES module were used as the background field, the wind fields obtained from tower anemometers and automatic weather station site supplementation were used as the observation field, and the corrected analysis field was used as the fused wind field.
[0043] The values are from the wind measurement towers and automatic weather stations (corresponding to the observed u-wind and v-wind), with a total of n observation stations. The influence radius R is taken as 12 kilometers, and the observation gap compensation value of the k-th observation station is... The distance to the specified grid point (i, j) is r. The weight function is obtained according to equation (3). . The background grid point values (corresponding to the u-wind and v-wind calculated by WRF-LES) are used. Due to the fine grid resolution, the grid point closest to the meteorological tower and automatic weather station is taken as the background value during the iterative calculation. The calculation is performed m times according to equation (4) (m is counted from 1). As the initial background field, (For the background field of the m-th iteration) Iteration, obtained The grid point values are the fused wind field values. Collect and organize the corresponding fused product datasets.
[0044] (3)
[0045] (4)
[0046] S3. Convolutional Neural Network Modeling and Near-Term Correction Forecast: A convolutional neural network model is established and trained between the forecast product dataset and the fused product dataset. In actual use, the forecast product of the WRF-LES module is input into the trained convolutional neural network model to obtain the corrected forecast wind field.
[0047] S31, Convolutional Neural Network Modeling
[0048] For the U-wind and V-wind forecasts from the WRF-LES module and fusion products within the same time period, normalization was performed at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters, respectively. The dataset was then divided into training, validation, and test sets in a 70%:15%:15% ratio.
[0049] The input layer of the convolutional neural network is the predicted U-wind and V-wind at different heights of the product (10m, 20m, 35m, 50m, 70m, 90m). The output layer is the fused U-wind and V-wind at the corresponding heights. The convolutional neural network is constructed through convolutional layers, activation functions, pooling layers, and fully connected layers. The mean squared error is used as the loss function and the Adam optimizer is used for model training to establish the convolutional neural network model.
[0050] S32, Near-surface wind field now-corrected forecast
[0051] In actual forecasting, based on the parameters of the convolutional neural network and combined with the characteristics of wind field continuity, the WRF-LES module is used to directly forecast the u-wind and v-wind for the next 0-6 hours. After normalization preprocessing, the forecasts are fed into the trained convolutional neural network model to further correct the direct forecast products of the WRF-LES module, resulting in corrected forecast wind fields at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters for the next 0-6 hours.
[0052] This invention obtains the local wind field profile by using the power law formula of near-surface wind field data from wind measurement towers and automatic weather stations. Then, the wind field is fused using WRF-LES. Based on this, convolutional neural network modeling is carried out to obtain model parameters, which are then used in WRF-LES real-time forecasting to obtain near-surface wind field correction forecasts.
[0053] This invention makes full use of local wind measurement tower and encrypted automatic weather station data, and integrates WRF-LES modeling, which can be used for local wind field forecasting. The corresponding algorithms and processes are applicable to wind field forecasting for wind power, low-altitude economy, etc.
[0054] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for local near-surface wind field fusion and nowcasting, characterized in that, The method includes the following steps: S1. Wind field gap filling above automatic weather stations: The wind shear index calculated by the wind measurement tower is used to fill the gap in the wind profile above the surrounding automatic weather stations. S2, Wind Field Fusion: Combine the forecast product dataset of WRF-LES large eddy simulation with the wind fields above tower wind measurement and automatic weather station sites to carry out Barnes stepwise correction wind field fusion to obtain the fused product dataset; S3. Convolutional Neural Network Modeling and Near-Term Correction Forecast: A convolutional neural network model is established and trained between the forecast product dataset and the fused product dataset. In actual use, the WRF-LES forecast product is input into the trained convolutional neural network model to obtain the corrected forecast wind field.
2. The method for local near-surface wind field fusion and nowcasting as described in claim 1, characterized in that, S1 includes: S11. Data quality control: Measure and acquire meteorological elements at the wind measurement height of each wind tower, remove wind field data during the period directly affected by the typhoon, and remove wind field data with wind speeds that deviate too much from the average wind speed. S12. Calculation of wind shear index at each level of the wind measuring tower: Based on the vertical temperature gradient, the wind field is included in three categories of samples: unstable, neutral, and stable according to atmospheric stability. Based on the power law formula of the vertical distribution of wind near the ground, the wind shear index of u wind and v wind under different stability conditions is calculated. u wind is east-west wind and v wind is north-south wind. S13. Filling in gaps in the wind profile from above the automatic weather station to a height of 90 meters: Centered on the anemometer tower, collect and organize real-time wind field observation data from automatic weather stations around the anemometer tower, and calculate the U-wind and V-wind from above the automatic weather station to a height of 90 meters based on the atmospheric stability and wind shear index of the anemometer tower.
3. The method for local near-surface wind field fusion and nowcasting as described in claim 2, characterized in that, S11 includes: The wind measurement tower has six layers at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters. Each layer is equipped with wind sensors, temperature sensors, and humidity sensors to measure wind speed, wind direction, temperature, and relative humidity, acquiring meteorological data every 10 minutes. Based on the China Meteorological Administration's optimal tropical cyclone path dataset, wind field data during the period of direct typhoon impact is removed. Using 50 meters as the standard layer, wind speed at 50 meters is measured. Calculate the mean of the sequence Standard deviation Further increase wind speed The wind field value was removed, and the corresponding data at the 10-meter, 20-meter, 35-meter, 70-meter, and 90-meter levels at the same time were also removed.
4. The method for local near-surface wind field fusion and nowcasting as described in claim 3, characterized in that, S12 includes: Select the temperatures T at the 20-meter and 70-meter levels of the wind measurement tower and calculate the vertical temperature gradient. ,according to For values less than 0, equal to 0, or greater than 0, atmospheric stability is classified into three categories: unstable, neutral, and stable. Based on changes in atmospheric stability, the wind field at the corresponding moment is included in the samples of the unstable, neutral, and stable categories. The wind field is horizontally decomposed into u-wind and v-wind. For the three types of samples—unstable, neutral, and stable—the power-law formula for the vertical distribution of near-surface wind is used, i.e., formulas (1) and (2), with 10 meters as the reference height. The corresponding known reference winds are respectively and Calculate the wind shear index of winds at altitudes u and v of 20m, 35m, 50m, 70m, and 90m under different atmospheric stability conditions. ,in, and Let z correspond to the u and v winds at altitudes of 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters, respectively; （1） （2） Calculate the wind shear index at different altitudes The average value is used as the wind shear index for different atmospheric stability at different altitudes (u and v).
5. The method for local near-surface wind field fusion and nowcasting as described in claim 4, characterized in that, S13 includes: Centered on the wind measuring tower, with a radius of 4 kilometers and distributed in a square, real-time wind field observation data from the dense automatic meteorological stations in the four directions of east, south, west and north of the wind measuring tower were collected and processed; according to the construction standards of automatic meteorological stations, the wind fields measured were all wind fields at a height of 10 meters, and the wind fields were decomposed into U wind and V wind; Atmospheric stability is calculated based on the temperature at 20 meters and 70 meters of the wind tower at the same time. Based on the unstable, neutral and stable atmospheric stability and wind shear index, the power law formulas of equations (1) and (2) are used to calculate the U-wind and V-wind at heights of 20 meters, 35 meters, 50 meters, 70 meters and 90 meters above each automatic station within 4 kilometers around the wind tower.
6. The method for local near-surface wind field fusion and nowcasting as described in any one of claims 2-4, characterized in that, S2 includes: S21. WRF-LES Design and Forecast: Using the 1KM output data of the WRF mode as the input data of the WRF-LES module, the parameters of the WRF-LES module are designed, and the corresponding forecast products are collected and organized into a forecast product dataset. S22. Fusion of wind fields from WRF-LES forecasts with tower anemometers and automatic weather station data: The Barnes stepwise correction method is adopted. The calculation results of the WRF-LES module are used as the background field, and the wind fields supplemented by tower anemometers and automatic weather station data are used as the observation field. The fused wind field is obtained after correction, and the fused product dataset is obtained.
7. The method for local near-surface wind field fusion and nowcasting as described in claim 6, characterized in that, S21 includes: using the 1KM output data of the WRF mode as the input data of the WRF-LES module; in the design of the WRF-LES module: the height distribution below 90 meters is set to 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters; the horizontal region of the model is processed by triple nesting, and the grid resolution ratio of the three nested regions from the outside to the inside is set to 25:5:1, the resolution of the innermost grid is set to 40 meters, and the innermost layer basically coincides with the 4-kilometer square area centered on the wind tower; the model start time is 05:00, 11:00, 17:00, and 23:00; the integration time is 3 hours, and the result is output once every 1 hour. Then, the U-wind and V-wind forecast products of the 3rd hour are taken, corresponding to 08:00, 14:00, 20:00, and 02:00 of the next day, respectively; the corresponding forecast products are collected and organized into a forecast product dataset.
8. The method for local near-surface wind field fusion and nowcasting as described in claim 7, characterized in that, S22 includes: The values are from the wind measurement tower and automatic weather station. There are a total of n observation stations; the influence radius R is taken as 12 kilometers, and the observation gap compensation value of the k-th observation station is... The distance to the specified grid point (i, j) is r. The weight function is obtained according to equation (3). ; The grid point values are the background field values, corresponding to the u-wind and v-wind calculated by the WRF-LES module. Since the grid point resolution is relatively fine, the grid point closest to the meteorological tower and the automatic weather station is taken as the background value during the iterative calculation. According to equation (4), m iterations are performed, with m counting backwards from 1. As the initial background field, The background field obtained in the m-th iteration is... The grid point values are the fused wind field values; collect and organize the corresponding fused product datasets: （3） （4）。 9. The method for local near-surface wind field fusion and nowcasting as described in claim 6, characterized in that, S3 includes: S31, Convolutional Neural Network Modeling For the U-wind and V-wind forecast products of WRF-LES module and fusion products in the same time period, normalization was performed at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters and 90 meters respectively, and the dataset was divided into training set, validation set and test set. The input layer of the convolutional neural network is the forecast of U-wind and V-wind at different heights of the product at 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters. The output layer is the fusion of the U-wind and V-wind at the corresponding heights. The convolutional neural network is constructed through convolutional layers, activation functions, pooling layers, and fully connected layers. The mean squared error is used as the loss function and the Adam optimizer is used to train the model and establish the convolutional neural network model. S32, Near-surface wind field now-corrected forecast In actual forecasting, based on the parameters of the convolutional neural network and combined with the characteristics of wind field continuity, the WRF-LES module is used to directly forecast the u-wind and v-wind for the next 0-6 hours. After normalization preprocessing, the forecasts are fed into the trained convolutional neural network model to further correct the direct forecast products of the WRF-LES module, resulting in corrected forecast wind fields at heights of 10 meters, 20 meters, 35 meters, 50 meters, 70 meters, and 90 meters for the next 0-6 hours.
10. The method for local near-surface wind field fusion and nowcasting as described in claim 9, characterized in that, In step S31, the dataset is divided into a training set, a validation set, and a test set in a ratio of 70%:15%:15%.