Launching site shallow wind speed forecast correction model construction method and system based on support vector machine

Abstract

The invention provides a launching site shallow wind speed forecast correction model construction method and system based on a support vector machine, and relates to the technical field of aerospace weather forecast guarantee. According to the method, a launching site 80m-height shallow wind forecasting correction model of different forecasting times and wind speed grades is established based on a support vector machine algorithm by utilizing site live data of a launching site and ECMWF element forecasting data, and the model can integrally improve the forecasting accuracy of 80m-height wind speed.

Description

Technical Field

[0001] The present invention relates to the technical field of aerospace weather forecasting and guarantee, and particularly relates to a method and system for constructing a correction model for shallow layer wind speed forecasting at a launch site based on a support vector machine. Background Art

[0002] Improving the refined forecasting level of wind speed, especially the wind speed at a height of 80 m, is of great significance for aerospace weather guarantee. For solid rockets, the whole layer of wind speed below 80 m must be less than the safe condition wind speed from the time when the wind protection bolts are removed before launch until the launch. For conventional liquid rockets, rocket assembly and satellite hoisting are both completed in front of the tower. To ensure the safety of personnel and equipment during the assembly process, it is necessary to ensure that the whole layer of wind speed below 80 m is below the safety threshold during operation. For liquid rockets performing manned missions, although the rocket and the spacecraft have been assembled in the general assembly building, during the process of vertically transporting the rocket-spacecraft combination to the tower, the rocket has not been fueled yet, and the combination is in a state of "top-heavy and bottom-light", and is very sensitive to the shallow layer wind speed at 80 m and below. When the wind speed is too high, it will cause excessive shaking of the rocket body, thus damaging the rocket body structure. Therefore, studying the wind speed forecasting deviation correction method and improving the accuracy of shallow layer wind forecasting at the launch site has important practical significance and application value.

[0003] Current numerical weather prediction (NWP) is the main means of wind field prediction. Affected by factors such as the physical process parameterization scheme and the accuracy of initial and boundary conditions, there are inevitable errors in the NWP predicted wind speed. Especially under complex underlying surface conditions, the wind speed prediction error is even greater. Therefore, correcting the bias of the wind speed predicted by NWP can effectively improve the prediction accuracy. With the rise of artificial intelligence (AI) technology, more and more studies have applied AI technology to various links of the prediction business. The numerical prediction interpretation application technology based on deep learning (DL) has shown significant advantages in mining the non-linear relationship of big data and spatial modeling. The combination of DL and NWP, namely DLWP (Deep Learning Weather Prediction), has become a research hotspot. Sun Quande et al. used three machine learning algorithms to correct the 10m wind speed in the North China region predicted by the numerical weather prediction model (European Centre for Medium-Range Weather Forecasts, ECMWF), and the correction results were better than the statistical output of the traditional model. Wang Zaiwen et al. used the support vector machine method to correct the element prediction results of the MM5 prediction system and applied it to the correction of the 2m air temperature and 10m wind speed prediction in the 2008 Olympic venues, and the prediction accuracy was significantly improved. There are many similar works on the correction of numerical model element predictions based on different DL methods. The above-mentioned large number of studies are mainly concerned with the 10m wind speed due to the needs of the application scenario. However, for nodes such as the vertical transfer of spacecraft, the hoisting and launching of rockets and satellites, the meteorological service pays more attention to the wind speed at a height of 80m and requires higher accuracy. Summary of the Invention

[0004] In view of the above technical problems, the present invention proposes a method and system for constructing a shallow-layer wind speed prediction correction model for a launch site based on a support vector machine.

[0005] The first aspect of the present invention discloses a method for constructing a shallow-layer wind speed prediction correction model for a launch site based on a support vector machine, and the method includes:

[0006] Step S1, extracting historical prediction data and site actual data of the target launch site; the historical prediction data includes the wind speed at a height of 10m, the wind speed at a height of 100m, the wind speed at 850hPa, the temperature at a height of 2m, and the temperature at 850hPa; the site actual data is the observed wind speed at a height of 80m.

[0007] Step S2, calculate the wind speed difference between 850 hPa and 10 m height and the temperature difference between 850 hPa and 2 m height, and based on the wind energy resource assessment standard of the wind farm, calculate the predicted wind speed at 80 m height according to the wind speed at 10 m height and the wind speed at 100 m height, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where, x i is the i-th input vector, and each input vector includes the wind speed at 10 m height, the wind speed at 100 m height, the wind speed at 850 hPa, the temperature at 2 m height, the temperature at 850 hPa, the predicted wind speed at 80 m height, the wind speed difference between 850 hPa and 10 m height, and the temperature difference between 850 hPa and 2 m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the on-site actual data;

[0008] Step S3, divide the sample set into n×h sample subsets according to different forecast times and wind speed levels; where, n is the number of wind speed levels and h is the forecast time;

[0009] Step S4, based on the support vector machine, construct n×h SVM models; each SVM model uses a radial basis kernel function to solve the optimal decision function;

[0010] Step S5, input each sample subset into a corresponding SVM model, and optimize the SVM model parameters by the cross-validation method. According to the principle of the minimum root mean square error of the predicted wind speed of the SVM model, screen and determine the penalty factor C and the kernel parameter δ value of each SVM model to obtain the correction model for the shallow-layer wind speed prediction at the launch site.

[0011] In the said Step S1, it further includes:

[0012] Screen and eliminate abnormal data.

[0013] In the said Step S2, the calculation formula for calculating the predicted wind speed at 80 m height is:

[0014]

[0015] where, v 80 is the predicted wind speed at 80 m height; v 100 and v 10 are the wind speed at 100 m height and the wind speed at 10 m height respectively; α is the wind shear index; z 100 、z 80 、z 10 are the position points at 100 m from the ground surface, 80 m from the ground surface and 10 m from the ground surface respectively.

[0016] In step S3, according to the requirements of the guarantee scenario, the wind speed at a height of 80m is divided into four levels: ≤6m / s, (6, 8]m / s, (8, 10]m / s, and >10m / s.

[0017] In step S3, the forecast times include 8 forecast times: 11:00, 14:00, 17:00, 20:00, 23:00 on the same day, 02:00, 05:00, and 08:00 on the next day.

[0018] The method further includes:

[0019] Step S6, constructing a test index, and calculating the probability of accurate forecast for each forecast time based on the calculation formula of the model wind speed forecast probability;

[0020] Among them, the test index is:

[0021] Within 24 hours of the forecast time, when the forecast wind speed ≤8m / s, the probability that the actual wind speed ≤10m / s is more than 85%;

[0022] Within 24 hours of the forecast time, when 8m / s < forecast wind speed < 10m / s, the probability that the actual wind speed ≤10m / s is more than 60%;

[0023] Within 24 hours of the forecast time, when the forecast wind speed ≥10m / s, the probability that the actual wind speed ≤8m / s is less than 20%.

[0024] The third aspect of the present invention discloses a system for constructing a correction model for shallow layer wind speed forecasting at a launch site based on a support vector machine. The system includes:

[0025] The first processing module is configured to extract historical forecast data and site actual data of the target launch site; the historical forecast data includes wind speed at a height of 10m, wind speed at a height of 100m, wind speed at 850hPa, temperature at a height of 2m, and temperature at 850hPa; the site actual data is the observed wind speed at a height of 80m;

[0026] The second processing module is configured to calculate the wind speed difference between 850hPa and 10m height and the temperature difference between 850hPa and 2m height, and based on the wind energy resource assessment standard of the wind farm, calculate the predicted wind speed at a height of 80m according to the wind speed at a height of 10m and the wind speed at a height of 100m, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where x iis the i-th input vector, and each input vector includes the wind speed at 10m height, the wind speed at 100m height, the wind speed at 850hPa, the temperature at 2m height, the temperature at 850hPa, the predicted wind speed at 80m height, the wind speed difference between 850hPa and 10m height, and the temperature difference between 850hPa and 2m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the on-site actual data of the station;

[0027] The third processing module is configured to divide the sample set into n×h sample subsets according to different forecast times and wind speed levels; where n is the number of wind speed levels and h is the forecast time;

[0028] The fourth processing module is configured to construct n×h SVM models based on the support vector machine; each SVM model uses a radial basis kernel function to solve the optimal decision function;

[0029] The fifth processing module is configured to input each sample subset into a corresponding SVM model, optimize the SVM model parameters through the cross-validation method, and screen and determine the penalty factor C and the kernel parameter δ value of each SVM model according to the principle of the minimum root mean square error of the wind speed predicted by the SVM model, so as to obtain the correction model for the shallow-layer wind speed forecast at the launch site.

[0030] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the method for constructing a correction model for the shallow-layer wind speed forecast at the launch site based on the support vector machine described in the first aspect of the present invention are implemented.

[0031] The fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in the method for constructing a correction model for the shallow-layer wind speed forecast at the launch site based on the support vector machine described in the first aspect of the present invention are implemented.

[0032] In summary, the solution proposed by the present invention has the following technical effects: By using the actual site data of the launch site and the ECMWF element forecast data, and based on the support vector machine algorithm, a correction model for the shallow-layer wind forecast at 80 m height of the launch site is established for different forecast times and wind speed levels. This model can overall improve the forecast accuracy of the wind speed at 80 m height, and the root mean square error of the shallow-layer wind speed forecast by the model at different forecast times is reduced to varying degrees compared with the ECMWF data. Specifically, in order to improve the forecast ability of the model for shallow-layer wind, on the one hand, n×h shallow-layer wind sub-models are constructed respectively for different wind speed intervals n and forecast time limits h, eliminating the interference of the daily variation of wind speed on the model; on the other hand, the dimension of the data sample is increased, considering the influence of the distribution of elements such as temperature, wind speed, and boundary layer height in the vertical direction on the 80 m wind field, making the simulation of the forecast process more realistic and closer to the actual situation, and being able to greatly reduce the forecast deviation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0034] Figure 1 Flowchart of a method for constructing a correction model for shallow-layer wind speed forecast at a launch site based on support vector machine according to an embodiment of the present invention;

[0035] Figure 2 Schematic diagram of the architecture of the correction model for shallow-layer wind speed forecast at the launch site according to an embodiment of the present invention;

[0036] Figure 3 Schematic diagram of the probability that the actual situation is ≤10 m / s when the model forecasts the shallow-layer wind speed at the launch site to be ≤8 m / s at different forecast times within the 24-hour forecast time limit according to an embodiment of the present invention;

[0037] Figure 4 Schematic diagram of the probability that the actual situation is ≤10 m / s when the model forecasts the shallow-layer wind speed at the launch site to be (8, 10] m / s at different forecast times within the 24-hour forecast time limit according to an embodiment of the present invention;

[0038] Figure 5 Schematic diagram of the probability that the actual situation is ≤8 m / s when the model forecasts the shallow-layer wind speed at the launch site to be >10 m / s at different forecast times within the 24-hour forecast time limit according to an embodiment of the present invention;

[0039] Figure 6(a)-(d) are schematic diagrams of the 80m wind speed correction results for each forecast time when the predicted wind speed in the embodiments of the present invention is ≤6m / s, (6, 8]m / s, (8, 10]m / s, and >10m / s respectively;

[0040] Figure 7 is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] The prior art provides a statistical method for depicting the similarity of weather processes. It transforms the forecasts arranged in chronological order into a similar space to find the historical forecast that is most similar to the current forecast (defined by a specific distance), that is, by defining an appropriate distance to measure the similarity between the historical forecast and the current forecast. This statistical method specifically includes the following steps: S1: Obtain the 70m height grid forecast products output by the WRF-based professional numerical wind energy forecast model and the hourly wind measurement data at 70m height of the wind farm anemometer tower; S2: Build a correction model by wind speed magnitude and season. The first two months of each season are the model construction period, and the last month is the model verification period; S3: Assign corresponding weights to the historical forecasts according to the similarity to the current forecast; S4: Perform weighted averaging on the observed values of the similar historical forecasts to obtain the corrected forecast.

[0043] Aviation meteorological support has a high demand for refined shallow-layer wind speed forecasts. For example, the sensitivity to wind in links such as satellite and rocket hoisting, rocket vertical transfer, and launch varies. The prior art divides the wind speed into 7 intervals of [0, 3), [3, 5), [5, 8), [8, 12), [12, 16), [16, 20), and [20, +∞)m / s based on the start-stop forecast requirements of wind turbines, which is not applicable to the field of aviation meteorological support.

[0044] By statistically analyzing the shallow-layer wind speed at the launch site, it is found that the shallow-layer wind speed has an obvious diurnal variation pattern. However, the prior art only establishes data sets for different wind speed grades and does not involve relevant indicators characterizing the diurnal variation of shallow-layer wind. Based on the actual support at the launch site, the applicability of this method in the shallow-layer wind forecast at the launch site is poor.

[0045] The local shallow-layer wind speed has strong non-linear characteristics, which are not only related to the horizontal distribution changes and time-varying changes of elements, but also related to the vertical distribution changes of elements. This method does not consider the influence of temperature and wind speed differences in the vertical direction on the shallow-layer wind speed.

[0046] In view of the technical problems existing in the above method, a method for constructing a correction model for shallow-layer wind speed prediction at the launch site based on a support vector machine is provided in the present invention. The method includes:

[0047] Step S1, extracting historical forecast data and site actual data of the target launch site; the historical forecast data includes the wind speed at a height of 10 m, the wind speed at a height of 100 m, the wind speed at 850 hPa, the temperature at a height of 2 m, and the temperature at 850 hPa; the site actual data is the observed wind speed at a height of 80 m.

[0048] 850 hPa refers to the near-surface layer, an expression of an isobaric surface, which is approximately 1500 meters above sea level in terms of altitude. The wind speed at a height of 80 m refers to the wind speed at a height of 80 meters from the ground surface.

[0049] Optionally, the ECMWF grid forecast data starting at 08:00 at the launch site from January 1, 2018 to October 26, 2023 is extracted as input. The elements include the 2 m temperature, 10 m U-wind (horizontal wind), 10 m V-wind (vertical wind), 100 m U-wind (horizontal wind), 100 m V-wind (vertical wind), 850 hPa temperature, and 850 hPa wind speed forecasted every 3 hours within 0 - 24 hours; the on-time observation data of the meteorological observation station at the launch site is extracted, and the elements include the 2-minute average wind speed, air temperature, relative humidity, etc.; the 2-minute average wind speed measured by the 80 m anemometer tower at the observation station is extracted.

[0050] The data is selected from January 1, 2018 to October 26, 2023, with better sample representativeness, improving the wind speed prediction ability of the model. ECMWF: European Centre for Medium-Range Weather Forecasts, an international weather forecasting research and operational agency.

[0051] Due to reasons such as incomplete downloaded data and failure to process file errors in a timely manner, there are defects in the ECMWF data, or the actual data is abnormal due to power outages and other reasons for the wind measurement equipment at the launch site. First, quality control such as data screening is required; to ensure the quality of the data set, invalid data, that is, data with missing actual values or forecast values, needs to be excluded. Only when both are valid are they recorded as valid data. The number of valid samples for each forecast time after quality control is shown in Table 1.

[0052] Table 1 Number of valid samples within different wind speed ranges for each forecast time

[0053]

[0054]

[0055] Step S2, calculate the wind speed difference between 850 hPa and 10 m height and the temperature difference between 850 hPa and 2 m height, and based on the wind energy resource assessment standard of the wind farm, calculate the predicted wind speed at 80 m height according to the wind speed at 10 m height and the wind speed at 100 m height, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where, x i is the i-th input vector, and each input vector includes the wind speed at 10 m height, the wind speed at 100 m height, the wind speed at 850 hPa, the temperature at 2 m height, the temperature at 850 hPa, the predicted wind speed at 80 m height, the wind speed difference between 850 hPa and 10 m height, and the temperature difference between 850 hPa and 2 m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the site actual data;

[0056] In this step, two data dimensions of the wind speed difference between 850 hPa and 10 m and the temperature difference between 850 hPa and 2 m are added to characterize the vertical difference. And by calculating the predicted wind speed at 80 m height, the alignment of the forecast and the actual data in height is achieved.

[0057] In the step S2, the calculation formula for the predicted wind speed at 80 m height is:

[0058]

[0059] where, v 80 is the predicted wind speed at 80 m height; v 100 and v 10 are the wind speed at 100 m height and the wind speed at 10 m height respectively; α is the wind shear exponent; z 100 、z 80 、z 10 are the position points at 100 m from the ground surface, the position point at 80 m from the ground surface, and the position point at 10 m from the ground surface respectively. The wind shear exponent is a parameter describing the variation characteristics of wind speed with height, indicating the variation of wind speed in the vertical plane perpendicular to the wind direction, and its magnitude is related to factors such as surface roughness, terrain, temperature, and wind speed.

[0060] Weather forecasting is not a simple mathematical statistic, which involves many problems of dynamics and thermodynamics. In the process of model training, the influence of the distribution of elements such as temperature, wind speed, and boundary layer height in the vertical direction on the 80 m wind field is considered, making the simulation of the forecasting process more realistic, closer to the actual situation, and capable of greatly reducing the forecasting deviation.

[0061] Step S3: Divide the sample set into n×h sample subsets according to different forecast times and wind speed levels; where n is the number of wind speed levels and h is the forecast time.

[0062] In step S3, the wind speed at 80m is divided into four levels: ≤6m / s, (6, 8]m / s, (8, 10]m / s, and >10m / s according to the requirements of the guarantee scenario.

[0063] In step S3, the forecast times include 8 forecast times: 11:00 on the same day, 14:00 on the same day, 17:00 on the same day, 20:00 on the same day, 23:00 on the same day, 02:00 on the next day, 05:00 on the next day, and 08:00 on the next day.

[0064] According to the statistical situation of the actual wind speed data in the launch site area and taking into account the needs of aerospace meteorological support, the 80m wind speed is divided into four levels: ≤6m / s, 6 - 8m / s, 8 - 10m / s, and ≥10m / s. A total of 8 time - series data with a 3 - hour forecast frequency within 24 hours are used to establish 32 sub - models of 4×8. The division is more detailed and scientific, so as to better ensure the implementation of the task, which has important practical significance and application value.

[0065] Step S4: Based on the support vector machine, construct n×h SVM models; each SVM model uses a radial basis kernel function to solve the optimal decision function; please refer to Figure 2 , which is the architecture composition diagram of the shallow - layer wind speed prediction correction model for the launch site. Taking the two samples that assume the wind speed at 100m height of the launch site predicted by ECMWF is 9m / s and 14m / s at 14:00 and 23:00 respectively as examples, it illustrates the architecture of the shallow - layer wind speed prediction correction model.

[0066] When dealing with problems such as function approximation or prediction, the support vector machine describes the problem as a convex optimization problem. It non - linearly maps the input sample points from the input space to a high - dimensional feature space, and then selects a loss function to solve the minimum value of the loss function in the high - dimensional feature space. Compared with traditional function approximation algorithms, the support vector machine has obvious advantages in preventing over - learning, operation speed, and result accuracy.

[0067] Each SVM model uses a non - linear mapping ψ to map the input training data to a high - dimensional feature space and constructs an optimal decision function in the feature space: f(x) = wΨ(x)+b, where: Ψ(x) is the mapping function; w is the weight vector, and its dimension is the dimension of the high - dimensional space; b ∈ R is the bias.

[0068] Solve the function estimation problem (that is, solve w and b) by minimizing the risk function: Where: ‖ω‖ is the describing function; с is the penalty coefficient, which is the penalty degree for samples exceeding the error ε. The larger с is, the greater the penalty for samples with training errors exceeding ε, and it is used for the compromise of the approximation error and the control of the model complexity; ξ and ξ* are the slack variables introduced considering the allowable fitting error; ε is the output error requirement of the regression function. The smaller ε is, the smaller the output error of the regression function and the higher the fitting accuracy.

[0069] The Lagrange multiplier method and the radial basis kernel function are used to solve the optimal decision function. By solving the equation, the estimation function of the optimal decision function of SVM can be obtained as: K(x, x i ) is the radial basis kernel function. The introduction of the kernel function enables the function to construct the optimal classification plane in the feature space, minimizing the total distance RM of the samples to the plane, thereby achieving fitting. The radial basis kernel function includes the kernel function parameter δ.

[0070] The penalty factor C and the kernel function parameter δ in the support vector machine combined prediction model are two hyperparameters that affect the performance. To improve the prediction accuracy of the model and avoid the subjective blindness of parameter selection, it is necessary to optimize the model parameters.

[0071] Step S5: Input each sample subset into a corresponding SVM model, and optimize the SVM model parameters through the cross-validation method. According to the principle of the minimum root mean square error of the predicted wind speed of the SVM model, screen and determine the penalty factor C and the kernel parameter δ values of each SVM model to obtain the correction model for the shallow-layer wind speed prediction at the launch site. The experiment of the shallow-layer wind prediction correction model based on SVM in this paper uses the libsvm toolbox

[19] . SVM needs to optimize the penalty factor C and the kernel parameter δ, and uses the cross-validation method to obtain the optimal parameters. The penalty factors and kernel parameters of each sub-model are shown in Table 2.

[0072] Table 2 Optimal penalty factor C and kernel parameter δ of each sub-model

[0073]

[0074]

[0075] The root mean square error of the predicted wind speed RMSE is the most commonly used performance metric in wind speed prediction. The smaller the RMSE, the more accurate the overall wind speed prediction. The formula is: Where K is the total number of samples, X k is the predicted wind speed value of the k-th sample, and y k is the actual wind speed value of the k-th sample.

[0076] The method also includes:

[0077] Step S6: Construct inspection indicators and calculate the probability of accurate prediction for each forecast time step based on the calculation formula of the model wind speed forecast probability;

[0078] Among them, the inspection indicators are:

[0079] Within a 24-hour forecast period, when the forecast wind speed ≤ 8 m / s, the probability that the actual wind speed ≤ 10 m / s is more than 85%;

[0080] Within a 24-hour forecast period, when 8 m / s < forecast wind speed < 10 m / s, the probability that the actual wind speed ≤ 10 m / s is more than 60%;

[0081] Within a 24-hour forecast period, when the forecast wind speed ≥ 10 m / s, the probability that the actual wind speed ≤ 8 m / s is less than 20%.

[0082] For a certain forecast time step t, the calculation formula of the model wind speed forecast probability In the formula, forecast(t, v) is the number of all forecast cases that meet the conditions (t, v) and the inspection indicators, and model(t, v) is the number of all model forecast cases that meet the conditions (t, v). According to FP(t, v), the probability of accurate prediction for each time step can be obtained.

[0083] Figure 3 For the 24-hour forecast period, when the model forecasts that the shallow-layer wind speed at the launch site ≤ 8 m / s at different forecast time steps, the probability that the actual situation ≤ 10 m / s. The abscissa is the forecast time step, and the ordinate is the probability.

[0084] Figure 4 For the 24-hour forecast period, when the model forecasts that the shallow-layer wind speed at the launch site is in the range of (8, 10] m / s at different forecast time steps, the probability that the actual situation ≤ 10 m / s. The abscissa is the forecast time step, and the ordinate is the probability.

[0085] Figure 5 For the 24-hour forecast period, when the model forecasts that the shallow-layer wind speed at the launch site > 10 m / s at different forecast time steps, the probability that the actual situation ≤ 8 m / s. The abscissa is the forecast time step, and the value is the forecast probability.

[0086] Please refer to Figure 6 , where the abscissa is the forecast time step, the left coordinate is RMSE, the right coordinate is the RMSE reduction rate, and the dashed box is the percentage reduction of RMSE after model correction compared to the ECMWF forecast. Generally, due to the surface properties of the gobi, a relatively stable inversion layer is formed above the launch site after 23:00. Statistics show that in the case of no disturbance, the actual wind speed of the near-surface layer at the launch site is mostly small most of the time after 23:00, while ECMWF often forecasts a larger wind speed. The model has learned this situation of ECMWF's false reporting and made a good correction.

[0087] The solution proposed by the present invention has the following technical effects: By using the actual site data of the launch site and the ECMWF element forecast data, and based on the support vector machine algorithm, a correction model for the shallow-layer wind forecast at 80m height of the launch site is established for different forecast times and wind speed levels. This model can overall improve the forecast accuracy of the wind speed at 80m height, and the root mean square error of the shallow-layer wind speed forecast by the model at different forecast times is reduced to varying degrees compared with the ECMWF data. Specifically, in order to improve the forecast ability of the model for the shallow-layer wind, on the one hand, n×h shallow-layer wind sub-models are constructed respectively for different wind speed intervals n and forecast times h, eliminating the interference of the daily variation of the wind speed on the model; on the other hand, the dimension of the data sample is increased, considering the influence of the distribution of elements such as temperature, wind speed, and boundary layer height in the vertical direction on the 80m wind field, making the simulation of the forecast process more realistic and closer to the actual situation, and greatly reducing the forecast deviation.

[0088] The third aspect of the present invention discloses a system for constructing a correction model for the shallow-layer wind speed forecast at the launch site based on the support vector machine. The system includes:

[0089] A first processing module, configured to extract the historical forecast data and the actual site data of the target launch site; the historical forecast data includes the wind speed at 10m height, the wind speed at 100m height, the wind speed at 850hPa, the temperature at 2m height, and the temperature at 850hPa; the actual site data is the observed wind speed at 80m height.

[0090] A second processing module, configured to calculate the wind speed difference between 850hPa and 10m height and the temperature difference between 850hPa and 2m height, and based on the wind energy resource assessment standard of the wind farm, calculate the predicted wind speed at 80m height according to the wind speed at 10m height and the wind speed at 100m height, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where, x i is the i-th input vector, and each input vector includes the wind speed at 10m height, the wind speed at 100m height, the wind speed at 850hPa, the temperature at 2m height, the temperature at 850hPa, the predicted wind speed at 80m height, the wind speed difference between 850hPa and 10m height, and the temperature difference between 850hPa and 2m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the actual site data.

[0091] A third processing module, configured to divide the sample set into n×h sample subsets according to different forecast times and wind speed levels; where, n is the number of wind speed levels, and h is the forecast time.

[0092] The fourth processing module is configured to construct n×h SVM models based on a support vector machine; each SVM model uses a radial basis kernel function to solve the optimal decision function;

[0093] The fifth processing module is configured to input each sample subset into a corresponding SVM model, optimize the SVM model parameters through cross-validation, and screen and determine the penalty factor C and kernel parameter δ values of each SVM model according to the principle of the minimum root mean square error of the predicted wind speed by the SVM model, so as to obtain the correction model for predicting the shallow-layer wind speed at the launch site.

[0094] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the method for constructing a correction model for predicting the shallow-layer wind speed at the launch site based on a support vector machine described in the first aspect of the present invention are implemented.

[0095] Figure 7 FIG. is a structural diagram of an electronic device according to an embodiment of the present invention. As Figure 7 shown, the electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, near field communication (NFC), or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, a touchpad, or a mouse, etc.

[0096] Those skilled in the art can understand that Figure 7 the structure shown in is only a structural diagram of a part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0097] The fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in the method for constructing a correction model for predicting the shallow-layer wind speed at the launch site based on a support vector machine described in the first aspect of the present invention are implemented.

[0098] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a correction model for shallow wind speed prediction at a launch site based on a support vector machine, characterized in that, The method includes: Step S1, extracting historical forecast data and site actual data of the target launch site; the historical forecast data includes wind speed at 10m height, wind speed at 100m height, wind speed at 850hPa, temperature at 2m height, and temperature at 850hPa; the site actual data is the observed wind speed at 80m height; Step S2, calculate the wind speed difference between 850 hPa and 10 m height and the temperature difference between 850 hPa and 2 m height, and based on the wind farm wind energy resource assessment standard, calculate the predicted wind speed at 80 m height according to the wind speed at 10 m height and the wind speed at 100 m height, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where, x i is the i-th input vector, and each input vector includes the wind speed at 10 m height, the wind speed at 100 m height, the wind speed at 850 hPa, the temperature at 2 m height, the temperature at 850 hPa, the predicted wind speed at 80 m height, the wind speed difference between 850 hPa and 10 m height, and the temperature difference between 850 hPa and 2 m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the site actual data; Step S3, dividing the sample set into n×h sample subsets according to different forecast times and wind speed levels; where n is the number of wind speed levels and h is the forecast time; Step S4, based on the support vector machine, constructing n×h SVM models; each SVM model uses a radial basis kernel function to solve the optimal decision function; Step S5, inputting each sample subset into a corresponding SVM model, and optimizing the SVM model parameters through the cross-validation method. According to the principle of the minimum root mean square error of the wind speed predicted by the SVM model, screening and determining the penalty factor C and kernel parameter δ values of each SVM model to obtain the correction model for the shallow layer wind speed forecast at the launch site.

2. The method according to claim 1, wherein In the said Step S1, it further includes: Screening and eliminating abnormal data.

3. The method according to claim 1, wherein In the said Step S2, the calculation formula for the predicted wind speed at 80m height is: Among them, v 80 is the predicted wind speed at a height of 80m; v 100 and v 10 are the wind speeds at heights of 100m and 10m respectively; α is the wind shear index; z 100 、z 80 、z 10 are the position points at 100m, 80m, and 10m from the ground surface respectively.

4. The method according to claim 1, wherein In the said Step S3, the wind speed at 80m height is divided into four levels: ≤6m / s, (6, 8]m / s, (8, 10]m / s, and >10m / s according to the requirements of the guarantee scenario.

5. The method according to claim 1, wherein In the said Step S3, the forecast times include 8 forecast times: 11:00 on the same day, 14:00 on the same day, 17:00 on the same day, 20:00 on the same day, 23:00 on the same day, 02:00 on the next day, 05:00 on the next day, and 08:00 on the next day.

6. The method according to claim 1, characterized in that, The method further includes: Step S6, constructing a test index, and calculating the probability of accurate forecast for each forecast time based on the calculation formula of the wind speed forecast probability of the model; Among them, the test index is: Within 24h of the forecast validity period, when the forecast wind speed ≤8m / s, the probability that the actual wind speed ≤10m / s is more than 85%; Within 24h of the forecast validity period, when 8m / s < forecast wind speed < 10m / s, the probability that the actual wind speed ≤10m / s is more than 60%; Within 24h of the forecast validity period, when the forecast wind speed ≥10m / s, the probability that the actual wind speed ≤8m / s is less than 20%.

7. A system for constructing a correction model for shallow layer wind speed prediction at a launch site based on a support vector machine, characterized in that, The system includes: The first processing module is configured to extract historical forecast data and site actual data of the target launch site; the historical forecast data includes wind speed at 10m height, wind speed at 100m height, wind speed at 850hPa, temperature at 2m height, and temperature at 850hPa; the site actual data is the observed wind speed at 80m height; The second processing module is configured to calculate the wind speed difference between 850 hPa and 10 m height and the temperature difference between 850 hPa and 2 m height, and calculate the predicted wind speed at 80 m height based on the wind energy resource assessment standard of the wind farm according to the wind speed at 10 m height and the wind speed at 100 m height, and then construct a sample set {(x1, y1), (x2, y2), …, (x t , y t )}; where x i is the i-th input vector, and each input vector includes the wind speed at 10 m height, the wind speed at 100 m height, the wind speed at 850 hPa, the temperature at 2 m height, the temperature at 850 hPa, the predicted wind speed at 80 m height, the wind speed difference between 850 hPa and 10 m height, and the temperature difference between 850 hPa and 2 m height; y i is the label corresponding to the i-th input vector; t is the number of samples; the label corresponding to the i-th input vector is the site actual data; The third processing module is configured to divide the sample set into n×h sample subsets according to different forecast times and wind speed levels; where n is the number of wind speed levels and h is the forecast time; The fourth processing module is configured to construct n×h SVM models based on the support vector machine; each SVM model uses a radial basis kernel function to solve the optimal decision function; The fifth processing module is configured to input each sample subset into a corresponding SVM model, optimize the SVM model parameters through the cross-validation method, and screen and determine the penalty factor C and the kernel parameter δ value of each SVM model according to the principle of the minimum root mean square error of the predicted wind speed by the SVM model, so as to obtain the correction model for the shallow-layer wind speed prediction at the launch site.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. When the processor executes the computer program stored in the memory, the steps in a method for constructing a correction model for shallow-layer wind speed prediction at a launch site based on a support vector machine according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in a method for constructing a correction model for shallow-layer wind speed prediction at a launch site based on a support vector machine according to any one of claims 1 to 6 are implemented.

Images