Wind speed prediction method and system for offshore observation area

By constructing the target WRF mode and BP neural network model, combining the actual sea measurement data and parameterization scheme, the problem of insufficient prediction accuracy of high wind speed in sparse observation areas in the sea is solved, improving prediction accuracy and reducing calculation costs, and enhancing extreme wind speeds Capture ability.

CN119090145APending Publication Date: 2024-12-06SHANGHAI INVESTIGATION DESIGN & RES INST CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411248758.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-06
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prediction accuracy of high wind speed in sparse observation areas at sea is insufficient, especially when extreme wind speed changes, the simulation error of the existing WRF mode is large, making it difficult to capture wind speed changes, and the computing resources and time cost are high.

Method used

Build the target WRF mode, by assimilating the ERA5 reanalytical data with the sea measured data, driving the initial WRF mode with parameterization scheme, using the BP neural network model to correct the simulated wind speed deviation, realize high spatial and temporal resolution wind speed prediction, and reduce calculation costs .

Benefits of technology

It improves the accuracy of forecasting of strong winds at sea, reduces computing resources and time costs, enhances the ability to capture extreme wind speeds, and corrects the system error of WRF mode.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090145A_ABST
    Figure CN119090145A_ABST
Patent Text Reader

Abstract

The invention provides a wind speed prediction method and system for an offshore observation area, and the method comprises the steps: constructing a target WRF mode to output a simulated wind speed, and constructing a BP neural network model for outputting a predicted wind speed deviation of the simulated wind speed in the target WRF mode; and calculating the predicted wind speed according to the simulated wind speed obtained in the target WRF mode and the predicted wind speed deviation. After the wind speed is simulated in the target WRF mode, wind speed simulation is performed in combination with a data decomposition method and a neural network model, so that the system error of the target WRF model is corrected, and the gale prediction accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of real sea environment forecasting, and in particular, to a method and system for predicting wind speed in an offshore observation area. Background Art

[0002] The planning and construction of offshore wind farm projects in my country are exploring farther and deeper deep sea areas. According to regulations, the site of an offshore wind power project needs to be more than 30 km offshore or more than 30 m deep, referred to as the "single 30" regulation. Strong winds at sea often cause waves and may trigger storm surges. They are one of the most serious natural disasters faced by offshore wind farms during the construction and operation and maintenance phases, and have a serious impact on offshore platform construction operations, wind turbine facilities, and wind turbine safety loads. Accurate sea surface wind forecasts can help wind power operators to reasonably formulate maintenance plans, help deploy disaster prevention measures in advance, and reduce damage to wind turbines and components caused by extreme disaster weather. Due to the limited means, difficulty, and high cost of observing wind resources in deep sea areas, as well as large spatial and temporal limitations, the lack of measured data has a significant impact on the evaluation and prediction of offshore wind speeds.

[0003] At present, numerical weather forecast is the most common method of wind energy forecasting, which has the advantages of quantitative, rapid and continuous, and can effectively make up for the limitations of traditional sea surface wind resource prediction. Numerical weather forecast belongs to physical process, which mainly considers the physical factors affecting the wind field, such as terrain characteristics, air pressure and ambient temperature, and the physical interaction laws between them to predict the wind field. At present, the most representative is the mesoscale numerical model (Weather Research and Forecast, WRF) developed by scientific research institutions and universities such as the US Environmental Prediction Center and the US National Center for Atmospheric Research. The WRF model simulates atmospheric phenomena by dividing the earth's atmosphere into horizontal grids and vertical layers, and solving atmospheric dynamics, thermodynamics and hydrological cycle equations through discrete equations. This model is widely used in wind resource simulation and wind speed prediction. The advantage of the WRF model is that the physical process is clear. Even if there are few observation data, high-resolution regional data can be constructed by characterizing local thermodynamic processes.

[0004] However, there are still great limitations in using the WRF model alone for wind speed forecasting. On the one hand, the WRF model requires multi-layer nesting for high-precision simulation, which consumes huge computing resources and time costs. On the other hand, for sparse observation areas with a lack of observation data and poor spatiotemporal continuity, it is difficult to use measured data to evaluate the downscaled products of the WRF model, and it is even more difficult to further optimize the parameterization scheme. In addition, the WRF model has a poor effect in capturing extreme wind speeds, especially when the wind speed changes dramatically. The WRF simulation error is large, and there is a certain systematic error between the measured data and the measured data. Summary of the invention

[0005] The purpose of this application is to provide a method and system for predicting wind speed in an offshore observation area to solve the problem of insufficient accuracy in predicting strong wind speed in sparsely observed areas at sea.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions:

[0007] A method for predicting wind speed in an offshore observation area comprises: constructing a target WRF model, wherein the construction process of the target WRF model comprises: assimilating ERA5 reanalysis data and measured data in an offshore target observation area to construct WRF driving field data; driving an initial WRF model with the WRF driving field data in combination with a parameterization scheme of the offshore target observation area, simulating downscaling of wind speed in the offshore target observation area, comparing the simulation results with the observed wind speeds of multiple offshore target observation areas, and obtaining a WRF optimal parameterization scheme according to the comparison results; determining a WRF model with high spatiotemporal resolution in the offshore target observation area, which is obtained by combining the WRF optimal parameterization scheme with a bidirectional nested grid encryption method, and the WRF model with high spatiotemporal resolution is the target WRF model; constructing a BP neural network model for outputting a predicted wind speed deviation of the target WRF model simulated wind speed; and calculating the predicted wind speed according to the simulated wind speed obtained by the target WRF model and the predicted wind speed deviation.

[0008] In some embodiments, the method of constructing a BP neural network model for outputting the predicted wind speed deviation of the target WRF model simulated wind speed includes: based on the measured wind speed WS obs The target WRF model simulates the same period wind speed WS WRF , calculate the simulated wind speed deviation X, X = WS obs -WS WRF ; Decompose the sequence composed of the simulated wind speed deviation X to obtain wind speed components X1~X n ;

[0009] The wind speed components X1 to X n The training set is used as input data to train the initialized BP neural network model to obtain the trained BP neural network model; the training set is {X t |t<<T}, where X t =(x t-m+1 , x t-m+2 , ..., x t-1 , x t ) is the input data of the initialized BP neural network, representing the historical data of the first m steps at time point t, and T represents the cutoff time step of the training data; the trained BP neural network model is used to output the predicted wind speed deviation It represents the predicted data for the hth time step after time point t.

[0010] In some embodiments, the wind speed of the WRF model to simulate the future period is set to WS WRF,fore , and the predicted wind speed deviation output by the trained BP neural network model Superimpose and calculate the predicted wind speed Right now

[0011] In some embodiments, when t<<T, the training set is used to predict future data; when t>T, the training set and part of the predicted values ​​are used to predict data for a longer period of time.

[0012] In some embodiments, the WRF driving field data is combined with the parameterization scheme of the offshore target observation area to drive the initial WRF model, the wind speed downscaling of the offshore target observation area is simulated, the simulation result is compared with at least the observed wind speed of the offshore target observation area, and the WRF optimal parameterization scheme is obtained according to the comparison result, including:

[0013] The observed wind speed of the offshore target observation area includes the reanalysis data, satellite remote sensing data or site observation data; the driving field data of the WRF is combined with a plurality of parameterization schemes of the selected offshore target observation area to drive the initial WRF model, so as to realize the downscaling simulation of the wind speed in the offshore target observation area, and obtain the simulated wind speed in the offshore target observation area; the root mean square error between the simulated wind speed and the observed wind speed in the offshore target observation area is calculated;

[0014] The formula for calculating the root mean square error is as follows:

[0015]

[0016] Among them, M i Indicates simulated wind speed, Station i Indicates site observation data, Remote i Represents satellite remote sensing data, Reanalysis i represents the reanalysis data, i represents the i-th grid point; a, b, and c represent the weights of the three root mean square errors, respectively; the parameterization scheme corresponding to the minimum root mean square error is the WRF optimal parameterization scheme.

[0017] In some embodiments, determining the WRF mode with high spatiotemporal resolution of the marine target observation area includes: using the WRF optimal parameterization scheme to perform two-layer bidirectional nesting on the marine target observation area, wherein the two-layer bidirectional nesting is that the mother area provides boundary conditions for the child area, and the calculation results of the child area are fed back to the mother area; the central longitude and latitude of the mother area are consistent with those of the child area; the grid point distance of the mother area is 9 km, and the number of grid points of the mother area is 22490;

[0018] The grid point distance of the sub-area is 3km, and the number of grid points in the sub-area is 75922; the WRF model with high temporal and spatial resolution in the offshore target observation area outputs 9km WRF model nested experimental simulation results and 3km WRF model nested experimental simulation results.

[0019] In some embodiments, the simulated wind speed and simulated direction output by the target WRF model are calculated by the following formula:

[0020]

[0021] WD=arctan(u / v)

[0022] Among them, u represents the latitudinal wind speed, v represents the radial wind speed, WS represents the wind speed, and WD represents the wind direction; the maximum wind speed value within a specific time period is selected as the strong wind speed forecast within that time period.

[0023] In some embodiments, the WRF optimal parameterization scheme is used to perform two-layer bidirectional nesting on the offshore target observation area, and the two-layer bidirectional nesting is that the parent area provides boundary conditions for the child area, and the calculation results of the child area are fed back to the parent area, and the WRF model nesting experimental simulation results of 1 km are output; the central longitude and latitude of the parent area are consistent with those of the child area; the grid point distance of the parent area is 3 km, and the number of grid points in the parent area is 75922; the grid point distance of the child area is 1 km, and the number of grid points in the child area is 250525.

[0024] In some embodiments, the 3 km WRF model nested experimental simulation results are interpolated to the 1 km scale using different interpolation methods, and different interpolation results are obtained accordingly. The interpolation results are expressed by the formula:

[0025]

[0026] Where Intep represents the interpolation result, WRF_1km represents the simulation result of the 1 km WRF model nested experiment, the subscript i represents the i-th grid point, and n represents the total number of grid points; the interpolation method corresponding to the minimum RMSE value is the optimal interpolation method, and the optimal interpolation method is combined with the 3 km WRF model nested experiment to obtain the target WRF model.

[0027] The second aspect of the present application provides a wind speed prediction system for an offshore observation area, comprising: a high temporal and spatial resolution WRF model construction module, a predicted wind speed deviation module and a wind speed prediction calculation module, wherein

[0028] The high spatiotemporal resolution WRF model construction module is used to construct a target WRF model. The construction process of the target WRF model includes: assimilating ERA5 reanalysis data and measured data in the offshore target observation area to construct WRF driving field data; driving the initial WRF model with the WRF driving field data in combination with the parameterization scheme of the offshore target observation area, simulating the downscaling of wind speed in the offshore target observation area, comparing the simulation results with the observed wind speeds of various offshore target observation areas, and obtaining the WRF optimal parameterization scheme according to the comparison results; determining the WRF model with high spatiotemporal resolution in the offshore target observation area, which is obtained by combining the WRF optimal parameterization scheme with bidirectional nested grid encryption, and the WRF model with high spatiotemporal resolution is the target WRF model; the predicted wind speed deviation module is used to construct a BP neural network model for outputting the predicted wind speed deviation of the target WRF model simulated wind speed; the wind speed prediction calculation module is used to calculate the predicted wind speed according to the simulated wind speed obtained by the target WRF model and the predicted wind speed deviation.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] The present application provides a wind speed prediction method and system for offshore observation areas, which constructs a target WRF model to output simulated wind speed. After the target WRF model simulates the wind speed, it combines a data decomposition method and a neural network model to simulate the strong wind speed, thereby correcting the system error of the target WRF model and improving the accuracy of strong wind prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1A schematic diagram showing a flow chart of a method for predicting wind speed in an offshore observation area according to an embodiment of the present application is shown;

[0033] Figure 2 A schematic diagram of the process of constructing a target WRF mode according to an embodiment of the present application is shown;

[0034] Figure 3 A schematic diagram of a two-layer bidirectional nested structure in an embodiment of the present application is shown;

[0035] Figure 4 A schematic diagram of the structure of a wind speed prediction system for an offshore observation area according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0036] The technical solutions of the specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings. These implementation methods are only used to illustrate the present application, but not to limit the present application.

[0037] The term "and / or" in this application is a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. In the present invention, the symbol " / " indicates that the associated objects are in an "or" relationship, for example, "A / B" means "A or B".

[0038] The terms "first" and "second" in the specification and claims of this application are used to distinguish different objects rather than to describe a specific order of the objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of the response messages.

[0039] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be considered as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0040] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc., and multiple elements refer to two or more elements, etc.

[0041] The following is a detailed description of a wind speed prediction method and system for an offshore observation area of ​​the present application in conjunction with the accompanying drawings.

[0042] Figure 1The following is a flow chart of a method for predicting wind speed in an offshore observation area according to an embodiment of the present application. The method for predicting wind speed in an offshore observation area according to an embodiment of the present application comprises the following steps:

[0043] like Figure 2 As shown, step S110, constructing a target WRF model, the construction process of the target WRF model is as follows: step S111, assimilating ERA5 (European Centre for Medium-Range Weather Forecasts Reanalysis v5) reanalysis data with sparse site measured data in the target observation area at sea, for example, data assimilation can be performed based on the Newton relaxation approximation (nudging) method to obtain a more accurate numerical weather forecast background field, obtain WRF driving long data, and use the assimilated numerical weather forecast background field as an initial condition to drive the WRF model for numerical weather forecasting. The Newton relaxation approximation method is a relatively mature four-dimensional assimilation technology currently developed. It is a continuous dynamic assimilation method, which adds a forcing function on the basis of the model control equation so that the model state gradually approaches the observation state. In one embodiment, the sparse site measured data in the target observation area at sea is subject to quality control, including format logic check, climate extreme range check, time consistency check, time continuity check, etc., and the measured data after quality control can be used for data assimilation.

[0044] Step S112, the WRF driving field data is combined with the parameterization scheme of the offshore target observation area to drive the initial WRF model, the wind speed downscaling of the offshore target observation area is simulated, the simulation results are compared with the observed wind speeds of various offshore target observation areas, and the WRF optimal parameterization scheme is obtained according to the comparison results. Specifically, based on the target study area, the parameterization scheme suitable for the corresponding area can be screened by consulting the literature, and a variety of alternative schemes are listed. The WRF driving field data obtained in the above steps is combined with different parameterization schemes to drive the WRF model to perform a one-year target area wind speed downscaling simulation. The above parameterization schemes can be, for example, Morrison 2-mom cloud microphysics scheme, Grell-Freitas cumulus convection scheme, New Goddard shortwave and longwave radiation scheme, YSU boundary layer scheme and Noah land surface process scheme. Downscaling simulation of the WRF model can obtain higher resolution and more detailed wind speed information. The simulation time length is set to one year, and the performance of the parameterization scheme on a long time scale can be evaluated, and the impact of seasonal and interannual changes on wind speed can be captured.

[0045] In one embodiment, the observed wind speeds in the aforementioned multiple offshore target observation areas may include reanalysis data, satellite remote sensing data, or site observation data. The aforementioned simulated wind speed is comprehensively compared with multi-source observation data such as reanalysis data, satellite remote sensing data, or site observation data, and the root mean square error between the simulated wind speed and the observed wind speed is calculated. The parameterization scheme corresponding to the minimum root mean square error is the WRF optimal parameterization scheme. The root mean square error calculation formula (1) is as follows:

[0046]

[0047] Among them, M i Indicates simulated wind speed, Station i Indicates site observation data, Remote i Represents satellite remote sensing data, Reanalysis i represents the reanalysis data, i represents the i-th grid point; a, b, and c represent the weights of the three root mean square errors, which are 0.6, 0.2, and 0.2 respectively.

[0048] For sparse marine observation areas with a lack of observation data and poor spatiotemporal continuity, it is difficult to use measured data to evaluate the downscaling products of the WRF model. The embodiment of the present application integrates multi-source data such as reanalysis data, satellite remote sensing data, and less site observation data, and optimizes the numerical model parameterization scheme through comprehensive scoring of multi-source data. The WRF model optimization process is realized by using comprehensive scoring of multi-source data.

[0049] Step S113, determine the WRF mode with high spatiotemporal resolution in the target observation area at sea, which is obtained by combining the WRF optimal parameterization scheme with the two-way nested grid encryption method. The WRF mode with high spatiotemporal resolution is the target WRF mode. Since the WRF mode uses equal longitude and latitude grids, it does not have the ability of local encryption. Using high-resolution grids throughout the field will consume a lot of computing resources and cannot meet the timeliness requirements. The use of nested encryption can achieve high resolution in key areas.

[0050] Therefore, combined with the WRF optimal parameterization scheme obtained in the previous steps, the WRF optimal parameterization scheme is used to perform a two-layer bidirectional nesting of the target area to obtain a WRF model with high temporal and spatial resolution. Bidirectional nesting is that the parent area provides boundary conditions to the child area, and the calculation results of the child area are fed back to the parent area. The second-layer nesting resolution is 9 kilometers and 3 kilometers respectively. In this way, a high temporal and spatial resolution forecast of the target area is achieved, that is, the forecast is updated every hour, and weather information is provided at a fine spatial scale of 3 to 9 kilometers. Figure 3As shown in the figure, the large box is the parent area, the small box is the child area, and the central longitude and latitude of the parent area and the child area are consistent, for example, 11.10°N and -73.52°E. The grid point distance of the parent area is set to 9 kilometers, and the number of grid points is 22490. The grid point distance ratio between the child area and the parent area is set to 3, and the number of grid points is 75922, so that the second-layer nesting resolution obtained after running WRF is 9 kilometers and 3 kilometers respectively.

[0051] The output of the target WRF model is the radial wind speed and zonal wind speed with a time resolution of 10 minutes and a height of 10 meters. The wind speed value is the square root of the sum of the squares of the radial wind and the zonal wind, as shown in formula (2); the wind direction is first calculated as the ratio of the zonal wind to the radial wind, and then calculated using the inverse tangent function, as shown in formula (3). Since the wind speed result output by the model is the average wind, the maximum wind speed value within a 1-hour time limit can be selected here to represent the gale forecast within the time limit, and the wind direction is the wind direction corresponding to this maximum wind speed value.

[0052]

[0053] WD=arctan(u / v)(3)

[0054] In the above formula, u represents the latitudinal wind speed, v represents the radial wind speed, WS represents the wind speed, and WD represents the wind direction.

[0055] Based on the 9km and 3km bidirectional nested target WRF models, a 1km nested experiment was conducted for some key areas to obtain the simulation results of the 1km nested target WRF model. However, although directly running the 1km WRF model nested experiment can obtain high-resolution results, the computational cost is high and it is not suitable for large-scale or long-term applications.

[0056] Therefore, the present application significantly reduces the computational cost while ensuring a certain accuracy through the interpolation method. In one embodiment, for some key marine target observation areas where the spatial resolution needs to be refined to 1 km, based on the 3 km nested target WRF model obtained in the above steps, different interpolation schemes such as inverse distance weighted method, bilinear interpolation method, Kriging interpolation method, radial basis function method, etc. are used to perform spatial interpolation, and the 3 km result is interpolated to 1 km scale.

[0057] The root mean square error (RMSE) between the results obtained by each interpolation scheme and the simulation results of the 1 km WRF model nesting experiment is calculated as shown in formula (4). The interpolation method with the smallest RMSE is selected as the optimal interpolation scheme. The optimal interpolation method is combined with the 3 km WRF model nesting experiment to obtain the target WRF model, thereby realizing rapid encryption simulation of key areas.

[0058]

[0059] In formula (4), Intep represents the interpolated result, WRF_1km represents the 1 km WRF model nested experimental simulation result, the subscript i represents the i-th grid point, and n represents the total number of grid points.

[0060] Traditionally, using the WRF model for kilometer-level simulation requires multiple layers of nesting, which consumes a lot of time and resources. This application compares the historical kilometer-level simulation of the WRF model with multiple interpolation schemes, selects the best interpolation scheme at a resolution of 1 kilometer, reduces computing resources and time costs, and realizes rapid encryption simulation of key areas.

[0061] Step S120: construct a BP neural network model for outputting the predicted wind speed deviation of the target WRF model simulated wind speed. obs The wind speed WS in the same period simulated by the target WRF model WRF , calculate the simulated wind speed deviation X, X = WS obs -WS WRF ; Use wavelet transform to decompose the sequence composed of simulated wind speed deviation X, and obtain the wind speed components X1~X n ; Establish an initialized BP neural network model and convert the wind speed components X1 to X n As input data, the initialized BP neural network model is trained to obtain the trained BP neural network model; the training set is {X t |t<<T}, where X t =(x t-m+1 , x t-m+2 , ..., x t-1 , x t ) is the input data for initializing the BP neural network, which represents the historical data of the first m steps at time point t, and T represents the cutoff time step of the training data; the trained BP neural network model is used to output the predicted wind speed deviation It is the model output data, which represents the predicted data of the hth time step after time point t.

[0062] Step S130, the predicted wind speed is calculated based on the simulated wind speed obtained by the target WRF mode and the predicted wind speed deviation. It can be understood that the target WRF mode in this step can be a nested WRF mode with a resolution of 3 kilometers, or it can be an upgraded optimized target WRF mode obtained by interpolating the 3-kilometer result to a 1-kilometer scale using an interpolation method. For the convenience of description, this application is expressed in terms of the target WRF mode.

[0063] Set the target WRF model to simulate the wind speed in the future period as WS WRF,fore , and the predicted wind speed deviation output by the trained BP neural network model Superimpose and calculate the predicted wind speed Right now The embodiment of the present application uses a time-rolling multi-step prediction method in the training and prediction process, that is, when t<<T, the training set is used to predict future data; when t>T, the training set and part of the predicted value are combined to predict data for a longer period of time, and so on, rollingly predicting future data, thereby improving the prediction accuracy.

[0064] Since the WRF model in the prior art has a poor effect in capturing extreme wind speeds, especially when the wind speed changes dramatically, the WRF model simulation error is large. This application adopts a data-driven and physical process hybrid model, that is, after simulating the wind speed with a high spatiotemporal resolution WRF model, it combines the data decomposition method and the neural network model to simulate the wind speed, thereby correcting the system error of the high spatiotemporal resolution WRF model model and improving the accuracy of strong wind prediction.

[0065] like Figure 4 As shown, the second aspect of the present application provides a wind speed prediction system for an offshore observation area, comprising: a high spatiotemporal resolution WRF model construction module, a predicted wind speed deviation output module and a wind speed prediction calculation module, wherein the high spatiotemporal resolution WRF model construction module is used to construct a target WRF model, and the construction process of the target WRF model includes: assimilating ERA5 reanalysis data and measured data in the offshore target observation area to construct WRF driving field data.

[0066] The WRF driving field data is combined with the parameterization scheme of the offshore target observation area to drive the initial WRF model, and the wind speed downscaling of the offshore target observation area is simulated. The simulation results are compared with the observed wind speeds of various offshore target observation areas, and the WRF optimal parameterization scheme is obtained according to the comparison results. In one embodiment, the observed wind speeds of the aforementioned various offshore target observation areas may include reanalysis data, satellite remote sensing data, or site observation data. Suitable parameterization schemes are selected from the literature, and the WRF driving field data obtained by the aforementioned module is combined with different parameterization schemes to drive the WRF model to perform a one-year target area wind speed downscaling simulation. The aforementioned simulated wind speed is comprehensively compared with multi-source observation data such as reanalysis data, satellite remote sensing data, or site observation data, and the root mean square error between the simulated wind speed and the observed wind speed is calculated. The parameterization scheme corresponding to the minimum root mean square error is the WRF optimal parameterization scheme. The root mean square error calculation formula (1) is as follows:

[0067]

[0068] Among them, M i Indicates simulated wind speed, Station i Indicates site observation data, Remotei Represents satellite remote sensing data, Reanalysis i represents the reanalysis data, i represents the i-th grid point; a, b, and c represent the weights of the three root mean square errors, which are 0.6, 0.2, and 0.2 respectively.

[0069] Determine the WRF model with high spatiotemporal resolution in the target observation area at sea, which is obtained by combining the WRF optimal parameterization scheme with the two-way nested grid encryption method. The WRF model with high spatiotemporal resolution is the target WRF model. Combined with the WRF optimal parameterization scheme obtained in the previous steps, the WRF optimal parameterization scheme is used to perform two-layer two-way nesting of the target area to obtain a WRF model with high spatiotemporal resolution. Two-way nesting is that the parent area provides boundary conditions to the child area, and the calculation results of the child area are fed back to the parent area. The second-layer nesting resolution is 9 kilometers and 3 kilometers, respectively. The grid point distance of the parent area is set to 9 kilometers, and the number of grid points is 22,490. The grid point distance ratio between the child area and the parent area is set to 3, and the number of grid points is 75,922, so that the second-layer nesting resolution obtained after running WRF is 9 kilometers and 3 kilometers, respectively. The output of the target WRF model is the radial wind speed and zonal wind speed with a time resolution of 10 minutes and a height of 10 meters. The wind speed value is the square root of the sum of the squares of the radial wind and the zonal wind, as shown in formula (2); the wind direction is first calculated as the ratio of the zonal wind to the radial wind, and then calculated using the inverse tangent function, as shown in formula (3). Since the wind speed result output by the model is the average wind, the maximum wind speed value within the 1 hour time limit can be selected here to represent the strong wind forecast within the time limit, and the wind direction is the wind direction corresponding to this maximum wind speed value.

[0070]

[0071] WD=arctan(u / v)(3)

[0072] In the above formula, u represents the latitudinal wind speed, v represents the radial wind speed, WS represents the wind speed, and WD represents the wind direction.

[0073] In one embodiment, a 1 km nesting experiment can be conducted for some key areas based on the 9 km and 3 km bidirectional nested target WRF models to obtain simulation results of the 1 km nested target WRF model. However, the experimental calculation cost of the 1 km resolution obtained by the previous method is relatively high. Therefore, the present application further adopts an interpolation method, based on the 3 km nested target WRF model obtained by the aforementioned module, and spatially interpolates the 3 km result to a 1 km scale through different interpolation schemes, such as the inverse distance weighted method, bilinear interpolation, Kriging interpolation, radial basis function method, etc.

[0074] The root mean square error (RMSE) between the results obtained by each interpolation scheme and the simulation results of the aforementioned 1 km WRF model nested experiment is calculated as shown in formula (4). The interpolation method with the smallest RMSE is selected as the optimal interpolation scheme, thereby achieving rapid intensification simulation of key areas.

[0075]

[0076] In formula (4), Intep represents the interpolated result, WRF_1km represents the 1 km WRF model nested experimental simulation result, the subscript i represents the i-th grid point, and n represents the total number of grid points.

[0077] The predicted wind speed deviation output module is used to construct a BP neural network model for outputting the predicted wind speed deviation of the target WRF model simulated wind speed. It can be understood that the target WRF model in this module can be a nested WRF model with a resolution of 3 kilometers, or it can be an upgraded optimized target WRF model obtained by interpolating the 3-kilometer result to a 1-kilometer scale using the interpolation method.

[0078] Specifically, based on the measured wind speed WS obs The wind speed WS in the same period simulated by the target WRF model WRF , calculate the simulated wind speed deviation X, X = WS obs -WS WRF ; Use wavelet transform to decompose the sequence composed of simulated wind speed deviation X, and obtain the wind speed components X1~X n ; Establish an initialized BP neural network model and convert the wind speed components X1 to X n As input data, the initialized BP neural network model is trained to obtain the trained BP neural network model; the training set is {X t |t<<T}, where X t =(x t-m+1 , x t-m+2 , ..., x t-1 , x t ) is the input data for initializing the BP neural network, which represents the historical data of the first m steps at time point t, and T represents the cutoff time step of the training data; the trained BP neural network model is used to output the predicted wind speed deviation It is the model output data, which represents the predicted data of the hth time step after time point t.

[0079] The wind speed prediction calculation module is used to calculate the predicted wind speed based on the simulated wind speed obtained by the target WRF model and the predicted wind speed deviation. Set the wind speed of the future period simulated by the WRF model to WS WRF,fore , and the predicted wind speed deviation output by the trained BP neural network model Superimpose and calculate the predicted wind speed Right now The embodiment of the present application uses a time-rolling multi-step prediction method in the training and prediction process, that is, when t<<T, the training set is used to predict future data; when t>T, the training set and part of the predicted value are combined to predict data for a longer period of time, and so on, rollingly predicting future data, thereby improving the prediction accuracy.

[0080] It should be understood that the specific process of each module executing the above corresponding steps has been described in detail in the above method embodiment, and for the sake of brevity, it will not be repeated here.

[0081] It should also be understood that the division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0082] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present application. These improvements and substitutions should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting wind speed in an offshore observation area, characterized in that: include: Constructing a target WRF model, wherein the construction process of the target WRF model includes: assimilating ERA5 reanalysis data and measured data in the offshore target observation area to construct WRF driving field data; driving the initial WRF model with the WRF driving field data in combination with the parameterization scheme of the offshore target observation area, simulating the downscaling of wind speed in the offshore target observation area, comparing the simulation results with the observed wind speeds in various offshore target observation areas, and obtaining the WRF optimal parameterization scheme according to the comparison results; determining a WRF model with high spatiotemporal resolution in the offshore target observation area, which is obtained by combining the WRF optimal parameterization scheme with bidirectional nested grid encryption, and the WRF model with high spatiotemporal resolution is the target WRF model; Constructing a BP neural network model for outputting a predicted wind speed deviation of the target WRF model simulated wind speed; The predicted wind speed is calculated based on the simulated wind speed obtained by the target WRF mode and the predicted wind speed deviation.

2. The method for predicting wind speed in an offshore observation area according to claim 1, characterized in that: The BP neural network model for outputting the predicted wind speed deviation of the target WRF mode simulated wind speed comprises: Based on the measured wind speed WS obs The target WRF model simulates the same period wind speed WS WRF , calculate the simulated wind speed deviation X, X = WS obs -WS WRF ; Decompose the sequence composed of the simulated wind speed deviation X to obtain wind speed components X1~X n ; The wind speed components X1 to X n The training set is used as input data to train the initialized BP neural network model to obtain the trained BP neural network model; the training set is {X t |t<<T}, where X t =(x t-m+1 ,x t-m+2 ,…,x t-1 , x t) is the input data of the initialized BP neural network, representing the historical data of the first m steps at time point t, and T represents the cutoff time step of the training data; The trained BP neural network model is used to output the predicted wind speed deviation It represents the predicted data for the hth time step after time point t.

3. The method for predicting wind speed in an offshore observation area according to claim 2, characterized in that: Set the wind speed of the WRF model to simulate the future period as WS WRF,fore , and the predicted wind speed deviation output by the trained BP neural network model Superimpose and calculate the predicted wind speed Right now 4. The method for predicting wind speed in an offshore observation area according to claim 3, characterized in that: When t<<T, the training set is used to predict future data; when t>T, the training set and part of the predicted values ​​are used to predict data for a longer period of time.

5. The method for predicting wind speed in an offshore observation area according to claim 1, characterized in that: The WRF driving field data is combined with the parameterization scheme of the offshore target observation area to drive the initial WRF model, and the wind speed downscaling of the offshore target observation area is simulated, and the simulation results are compared with at least the observed wind speed of the offshore target observation area, and the WRF optimal parameterization scheme is obtained according to the comparison results, including: The observed wind speed in the offshore target observation area includes the reanalysis data, satellite remote sensing data or site observation data; The WRF driving field data is combined with multiple parameterization schemes of the selected offshore target observation area to drive the initial WRF model, so as to realize the downscaling simulation of the wind speed in the offshore target observation area and obtain the simulated wind speed in the offshore target observation area; Calculating the root mean square error between the simulated wind speed and the observed wind speed in the target observation area at sea; The formula for calculating the root mean square error is as follows: Among them, M i Indicates simulated wind speed, Station i Indicates site observation data, Remote i Represents satellite remote sensing data, Reanalysis i represents the reanalysis data, i represents the i-th grid point; a, b, and c represent the weights of the three root mean square errors, respectively; The parameterization scheme corresponding to the minimum root mean square error is the WRF optimal parameterization scheme.

6. The method for predicting wind speed in an offshore observation area according to claim 5, characterized in that: Determine the WRF model with high temporal and spatial resolution for the target observation area at sea, including: Using the WRF optimal parameterization scheme to perform two-layer bidirectional nesting on the offshore target observation area, the two-layer bidirectional nesting is that the parent area provides boundary conditions for the child area, and the calculation results of the child area are fed back to the parent area; The central longitude and latitude of the parent area are consistent with those of the child area; The grid point distance of the mother region is 9 km, and the number of grid points in the mother region is 22490; The grid point distance of the sub-region is 3 km, and the number of grid points in the sub-region is 75922; The WRF model with high temporal and spatial resolution in the offshore target observation area outputs 9km WRF model nested experimental simulation results and 3km WRF model nested experimental simulation results.

7. The method for predicting wind speed in an offshore observation area according to claim 6, characterized in that: The simulated wind speed and simulated direction output by the target WRF model are calculated by the following formula: WD=arctan(u / v) Among them, u represents the zonal wind speed, v represents the radial wind speed, WS represents the wind speed, and WD represents the wind direction; The maximum wind speed value within a specific time period is selected as the maximum wind speed forecast within that time period.

8. The method for predicting wind speed in an offshore observation area according to claim 7, characterized in that: Use the WRF optimal parameterization scheme to perform two-layer bidirectional nesting on the offshore target observation area, wherein the two-layer bidirectional nesting is that the parent area provides boundary conditions for the child area, the calculation results of the child area are fed back to the parent area, and the 1km WRF model nesting experimental simulation results are output; The central longitude and latitude of the parent area are consistent with those of the child area; The grid point distance of the mother region is 3 km, and the number of grid points in the mother region is 75922; The grid point distance of the sub-area is 1 km, and the number of grid points in the sub-area is 250525.

9. The method for predicting wind speed in an offshore observation area according to claim 8, characterized in that: The target WRF model construction process also includes: The 3km WRF model nested experimental simulation results are interpolated to the 1km scale through different interpolation methods, and different interpolation results are obtained. The interpolation results are expressed by the formula: Intep represents the interpolated result, WRF_1km represents the 1 km WRF model nested experimental simulation result, subscript i represents the i-th grid point, and n represents the total number of grid points; The interpolation method corresponding to the minimum RMSE value is the optimal interpolation method. The optimal interpolation method is combined with the 3 km WRF model nesting experiment to obtain the target WRF model.

10. A wind speed prediction system for an offshore observation area, characterized in that: include: High temporal and spatial resolution WRF model building module, predicted wind speed deviation module and wind speed prediction calculation module, among which, The high spatiotemporal resolution WRF model construction module is used to construct a target WRF model. The construction process of the target WRF model includes: assimilating ERA5 reanalysis data and measured data in the offshore target observation area to construct WRF driving field data; driving the initial WRF model with the WRF driving field data in combination with the parameterization scheme of the offshore target observation area, simulating the downscaling of wind speed in the offshore target observation area, comparing the simulation results with the observed wind speeds in various offshore target observation areas, and obtaining the WRF optimal parameterization scheme according to the comparison results; determining the WRF model with high spatiotemporal resolution in the offshore target observation area, which is obtained by combining the WRF optimal parameterization scheme with bidirectional nested grid encryption. The WRF model with high spatiotemporal resolution is the target WRF model; The predicted wind speed deviation output module is used to construct a BP neural network model for outputting the predicted wind speed deviation of the target WRF mode simulated wind speed; The wind speed prediction calculation module is used to calculate the predicted wind speed according to the simulated wind speed obtained by the target WRF mode and the predicted wind speed deviation.

Citation Information

Patent Citations

  • Method for improving wind speed forecast accuracy

    CN103793511A

  • Value weather forecast and artificial intelligence (AI) coupling predication method for wind speed extremum of costal typhoon

    CN108983320A

  • Wind resource prediction method based on WRF and CNN convolutional neural network

    CN112100922A

  • Wind speed prediction method and system based on WRF mode sensitivity parameter adjustment

    CN116307018A

  • Plasma processing apparatus

    KR102902271B1