New energy wind field multi-scale prediction method and system based on large-scale analysis constraint

Abstract

The invention discloses a new energy wind field multi-scale prediction method and system based on large-scale analysis constraint, and belongs to the technical field of new energy prediction. The method comprises the steps of WRF configuration and physical process parameterization, establishment of an initial field and a boundary field, calculation of a background forecast field, assimilation of observation data, updating of a set disturbance field, calculation of a set average and a set covariance, introduction of large-scale constraints, generation of a mixed analysis field by using a mixed data assimilation system, generation of a new initial field and generation of a new initial disturbance field. And circularly forecasting until the forecasting is completed. The method at least has the advantages that high-precision meteorological data support can be provided for wind power plant power prediction, offshore wind energy resource evaluation and the like through accurate simulation and prediction of a wind field, especially a deep sea wind field, the accuracy of power prediction and wind energy resource evaluation is improved, a wind energy resource development and utilization strategy is optimized, and the development and utilization efficiency is improved. And the utilization of diversified energy and the development of an energy storage technology are promoted.

Description

Technical Field

[0001] This invention belongs to the field of new energy prediction technology, and more specifically, relates to a multi-scale prediction method and system for new energy wind farms based on large-scale analysis constraints. Background Technology

[0002] Wind farm forecasting plays a crucial role in power grid operation, especially in wind power. It provides effective reference for wind power load forecasting, power system dispatching, renewable energy planning, power equipment planning and selection, and disaster prevention and response, which helps to achieve efficient operation of the power system and reliable power supply. The accuracy of wind farm forecasting directly affects the effectiveness of its application.

[0003] Currently, high-precision numerical weather prediction (NMR) technology is widely used both domestically and internationally for wind field forecasting. Accurate NMR forecasting hinges on the initial field and boundary conditions. Data assimilation is an effective method for fully utilizing all available observational data to accurately represent the current atmospheric state. It significantly improves the quality of data analysis and the effectiveness of numerical weather prediction. Furthermore, data assimilation can incorporate meteorological satellite observations, compensating for the lack of observations in marine, especially deep-sea, regions. Mainstream data assimilation methods include 3DVar (Three-Dimensional Variational Data Assimilation), EnKF (Ensemble Kalman Filter), and En3DVar (Ensemble Three-Dimensional Variational Data Assimilation).

[0004] The 3DVar assimilation method is widely used in wind field forecasting due to its ease of implementation and operation. However, its drawback is that the background error covariance is static, which limits the analysis quality. The EnKF assimilation method can obtain the flow-dependent background error covariance reflecting the wind field through ensemble forecast error statistics. However, its drawback is that the background error covariance includes noise from long-range covariances due to the limited number of ensemble members. The En3DVar method combines the advantages of both methods, introducing the "flow-dependent" ensemble background error covariance into the variational cost function. It not only possesses the quality control and equipotential of direct assimilation of unconventional data from variational methods, but also includes the "flow-dependent" ensemble covariance that describes the evolution of weather processes, providing an optimal solution for wind field forecasting.

[0005] The atmosphere is a unified whole. Global large-scale wind fields provide the background wind fields for local wind fields, influencing their formation and evolution. Conversely, local wind fields also exert feedback effects on large-scale wind fields. Based on this relationship, scholars both domestically and internationally have attempted to utilize large-scale wind field information from global models to improve regional model predictions. This involves restarting the data assimilation cycle to reintroduce large-scale information and eliminate accumulated errors in regional models. However, a drawback is that some periodic methods may lose small-scale information established in previous prediction cycles, affecting prediction accuracy. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a multi-scale prediction method and system for new energy wind fields based on large-scale analysis constraints. This method employs the En3DVar hybrid data assimilation framework, assimilates satellite wind observation data into wind field prediction, and adds global model information as an additional observation value to the En3DVar variational cost function of hybrid data assimilation. By introducing large-scale analysis constraints, the prediction of wind fields is improved.

[0007] The present invention adopts the following technical solution. The first aspect of the present invention provides a multi-scale prediction method for new energy wind farms based on large-scale analysis constraints, comprising the following steps:

[0008] Configure the mesoscale weather numerical forecasting model WRF model and parameterize the physical processes; use real-time forecast data from NCEP, interpolate the initial field through the WPS initialization module to obtain the initial field, and use GFS data to establish the lateral boundary conditions;

[0009] Starting from the NCEP analysis values, integrate over several hours to obtain several hours of RMSE perturbation profiles, generate K random perturbations, perform several hours of ensemble forecasting, and obtain a background forecast field of K members;

[0010] Read conventional observation data and satellite wind observation data, and assimilate them; use K-member background forecast field combined with observation data, and use ETKF transformation to update the ensemble perturbation field;

[0011] Based on the ensemble forecast results obtained from several hours of ensemble forecasts, the ensemble mean and ensemble covariance are calculated; large-scale analysis constraints are introduced into the cost function of En3Dvar, the ensemble mean is updated based on the ensemble mean and ensemble covariance, and combined with observation data, a mixed analysis field is generated.

[0012] The updated ensemble perturbation field is superimposed onto the hybrid analysis field to form a new set of ensemble members, which serves as the new initial field. A new initial perturbation field is generated, and the ensemble forecast enters the next cycle. The process is iteratively executed until the forecast period ends.

[0013] Preferably, the configuration of the mesoscale weather numerical forecasting model (WRF model) includes:

[0014] Select the forecast model and data assimilation system; and use a two-layer nested grid with a horizontal grid spacing of 5 km, 45 vertical layers, and a 50 hPa model top;

[0015] The parameterization of the physical process includes:

[0016] Radiation processes, microphysical processes, boundary layer schemes, cumulus convection parameterization, and road surface processes were studied using the WSM6 microphysical scheme, the RRTMGT radiative transfer scheme, the Tiedtke cumulus convection scheme, the YSU planetary boundary layer scheme, the modified MM5 surface layer scheme, and the Noah land surface model scheme.

[0017] Preferably, the conventional observation data comes from GDAS provided by NCEP, including: SYNOP, METAR, SHIP+BUOY, GTS radiosonde observation data, AMDAR, and SATWND;

[0018] The satellite-observed winds include: cloud-guided winds, sea surface winds from China's Wind Catcher-1 A / B satellites, and atmospheric motion vectors based on radiance from Japan's Himawari-8 meteorological satellite.

[0019] Preferably, the assimilation of conventional observation data and satellite observation wind includes:

[0020] Wind observation data from different sources and in different formats are uniformly organized into a format suitable for assimilation processing;

[0021] Denoising, missing value imputation, and / or data smoothing algorithms are used to control the quality of observational data.

[0022] Error propagation analysis and uncertainty assessment are used to evaluate and analyze the errors in the observation data, and the errors are allocated according to their sources and nature.

[0023] Preferably, NCEP GFS data is used, and the initial field is obtained by interpolation through the WPS initialization module in continuous loop En3DVar operation. The data is assimilated every 12 hours, and the lateral boundary conditions are provided by GFS reanalysis data every 6 hours.

[0024] Integrating the NCEP analysis values ​​over 6 hours yields the 6-hour RMSE perturbation profile, generating 30 random perturbations. A 6-hour ensemble forecast is then performed to obtain the background forecast field for 30 members.

[0025] Preferably, the cost function for incorporating large-scale analysis constraints into En3Dvar includes:

[0026] A pre-adjustment transformation is used to transform the original cost function of the En3DVar method, converting the calculation of the original cost function and its gradient into an iterative solution of the new control variables;

[0027] Large-scale analysis constraints are added to the transformed cost function to incorporate them into the assimilation of hybrid En3DVar data.

[0028] Preferably, the large-scale analysis constraint term is expressed by the following formula:

[0029]

[0030] In the formula:

[0031] This refers to meteorological data derived from global model analysis or forecasts and assimilated into large-scale observational data.

[0032] H c This represents a linear operator that maps variables from the grid space of a region model to their locations in a large-scale field.

[0033] R c This represents the error covariance matrix of large-scale analysis of meteorological data.

[0034] x′ represents the analytical increment in the data assimilation system after introducing the ensemble forecast covariance matrix of “flow correlation”.

[0035] This indicates a large-scale background field forecast.

[0036] Preferably, the step of updating the ensemble mean based on the ensemble mean and ensemble covariance, and generating a mixed analysis field by combining observational data, includes:

[0037] The introduction of large-scale analysis constraints into the assimilation of hybrid En3DVar data includes:

[0038] Read background field grid data x b And related background error covariance matrix B statistical data;

[0039] Read the predicted perturbation values ​​of m set members

[0040] Reading various types of observation data, including satellite observations. o Large-scale observation data And set the corresponding observation error;

[0041] Calculate the innovation vector y in the cost function o -H(x b ),

[0042] Set the initial values ​​for the analysis control variable (v, w) during iteration, where w is composed of m vector fields w1, w2, ..., w m composition;

[0043] Calculate the cost function value based on the control variable (v,w), and then calculate the background term value, analysis increment value x′, observation term value, and large-scale analysis constraint term value in the cost function in sequence.

[0044] Calculate the gradient of the cost function with respect to the control variables (v, w);

[0045] Based on the obtained gradient values, calculate the gradient descent direction and optimal step size, and update the iterative values ​​of the control variables (v, w) until the convergence criterion is reached;

[0046] The final analytical value is obtained by incremental calculation of background field forecast overlay analysis.

[0047] Preferably, the vertical mode layer index for initiating large-scale analysis constraints is 10, and data refinement is performed over a distance of 60 kilometers.

[0048] The En3DVar module uses 75% ensemble covariance and 25% static covariance, with a localization scale of 150km.

[0049] The static background error covariance matrix is ​​calculated using the NMC method.

[0050] A second aspect of this invention provides a multi-scale prediction system for new energy wind farms based on large-scale analysis constraints, which runs the aforementioned multi-scale prediction method for new energy wind farms based on large-scale analysis constraints. The system is characterized by including: a WRF platform and a data acquisition module.

[0051] The data acquisition module is used to acquire conventional observation data and satellite observation data, and input them into the WRF platform for assimilation;

[0052] The WRF platform incorporates large-scale analysis constraints into the cost function of En3Dvar to generate a hybrid analysis field for iterative iteration until the forecast period ends.

[0053] Compared with existing technologies, the beneficial technical effects of this invention include at least the following: This invention proposes a multi-scale prediction method and system for new energy wind farms based on large-scale hybrid variational analysis constraints. This method adopts the En3DVar hybrid data assimilation framework, assimilates satellite wind observation data into wind farm prediction, and adds global model information as an additional observation value to the hybrid data assimilation En3DVar variational cost function. By introducing large-scale analysis constraints, a more accurate simulation and prediction field of wind farms, especially deep-sea wind farms, is generated. This provides high-precision meteorological data support for operational scenarios such as wind farm power prediction, offshore wind energy resource assessment, and power meteorological disaster prediction and early warning, thereby improving the accuracy of power prediction, wind energy resource assessment, and power meteorological disaster prediction. Attached Figure Description

[0054] Figure 1 Flowchart of En3DVar data assimilation to introduce large-scale analysis constraints. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0056] like Figure 1 As shown, Embodiment 1 of the present invention provides a multi-scale prediction method for new energy wind fields based on large-scale analysis constraints, including: WRF configuration and physical process parameterization, establishing initial and boundary fields, calculating background forecast fields, observation data assimilation, updating ensemble perturbation fields, calculating ensemble averages and ensemble covariance, generating a mixed analysis field using a mixed data assimilation system, generating a new initial field and a new initial perturbation field, and iteratively forecasting until the forecast is complete. This generates more accurate simulation and prediction fields for wind fields, especially deep-sea wind fields.

[0057] The improvements made by the method to conventional prediction technology for new energy wind fields include at least the following: First, fully leveraging the advantages of satellite observation data, such as its wide coverage and continuous observation, to solve the problem of insufficient observation data at sea, especially in the deep sea, by assimilating satellite wind observation data; Second, adding global model information as an additional observation value to the En3DVar variational cost function of the mixed data assimilation, thereby improving wind field prediction by introducing large-scale analysis constraints.

[0058] The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints specifically includes the following steps:

[0059] Step 1: Configure the mesoscale weather numerical forecasting model WRF and parameterize the physical processes.

[0060] Preferably, but not limitingly, the configuration of the mesoscale weather numerical forecasting model WRF model includes: using the next-generation WRF-ARWV4.3 (http: / / www2.mmm.ucar.edu / wrf / users) as the forecasting model, using WRFDAV4 (https: / / www2.mmm.ucar.edu / wrf / users / wrfda / index.html) as the data assimilation system, with two nested grids, a horizontal grid spacing of 5 km, 45 vertical layers, and a 50 hPa model top, generated on a computational domain in the Pacific Northwest.

[0061] Preferably, but not limitingly, the physical process parameterization, including radiation processes, microphysical processes, boundary layer schemes, cumulus convection parameterization, and road surface processes, employs the WSM6 microphysical scheme (WRF Single-Moment 6-class), the RRTMGT radiative transfer scheme (Rapid Radiative Transfer Model for General circulation model applications), the Tiedtke Cumulus Parameterization Scheme, the YSU Planetary Boundary Layer Scheme, the modified MM5 surface layer scheme, and the Noah-MP land surface model scheme (Noah-MultiParameterization LandSurface Model).

[0062] Step 2: Using real-time forecast data from NCEP (www.nco.ncep.noaa.gov / pmb / products / gfs), i.e. GFS (Global Forecast System) data, the initial field is obtained by interpolation through the WPS (WRF Preprocessing System) initialization module, and the lateral boundary conditions are established using GFS data.

[0063] Preferred but not limited, NCEP GFS data is used, interpolated by the WPS initialization module, and the initial field is obtained in a continuous loop of En3DVar operation, assimilated once every 12 hours, and the lateral boundary conditions are provided by GFS reanalysis data once every 6 hours.

[0064] Step 3: Integrate the NCEP analysis values ​​over several hours to obtain the RMSE perturbation profile over several hours, generate K random perturbations, perform ensemble forecasts over several hours, and obtain the background forecast field of K members.

[0065] Preferredly but not restrictively, the 6-hour RMSE perturbation profile is obtained by integrating from the NCEP analysis value, generating 30 random perturbations, and performing a 6-hour ensemble forecast to obtain a background forecast field with 30 members.

[0066] In a preferred but non-limiting embodiment of the present invention, the initial set members are obtained by perturbation using the RandomCV method based on the GFS reanalysis field, and are updated using the ETKF (Ensemble Transform Kalman Filter) method in subsequent assimilation cycles.

[0067] Step 4: Read conventional observation data and satellite wind observation data, and assimilate them. For example, but not limited to, for conventional wind observations and satellite wind observations, use the observation preprocessing module in WRFDA (Weather Research and Forecasting Model Data Assimilation, WRF data assimilation) to perform data processing, quality control and error allocation.

[0068] Preferred but not limited, conventional data primarily comes from GDAS (Global Data Assimilation System, www.emc.ncep.noaa.gov / gmb / gdas) observation data provided by NCEP, including: SYNOP (Surface Synoptic Observations), METAR (Meteorological Terminal Aviation Routine Weather Report), SHIP+BUOY (Ship and Buoy Observations), Global TTS (Global Telecommunication System) radiosonde observation data (Rawinsonde), AMDAR (Aircraft Meteorological Data Relay), SATWND (Satellite Wind), and other conventional observation data.

[0069] Preferred but not limited, the satellite-observed wind, in addition to cloud-guided wind, also assimilates sea surface winds from China's Fengyun-1 A / B satellites and atmospheric motion vectors from the radiance of Japan's Himawari-8 meteorological satellite.

[0070] It is worth noting that the types of observational data involved in assimilation given above are a preferred but non-limiting implementation of the present invention. In the prior art, the types of observational data may be different. After research by the inventors, the above-mentioned types of observational data are a preferred way of implementing the present invention. However, those skilled in the art can choose more or fewer types of observational data, all of which fall within the scope of the present invention.

[0071] Preferred but not limited, the process of running WRFDA for observation preprocessing mainly includes: data processing, quality control, and error allocation of observation data through the observation preprocessing module in WRFDA.

[0072] Further optimization, but not limitation, of the observation data includes data processing, quality control, and error allocation, specifically including:

[0073] Step 4.1: Organize wind observation data from different sources and in different formats into a format suitable for assimilation.

[0074] Step 4.2: Use a series of algorithms and techniques such as denoising, filling in missing values, and data smoothing to control the quality of observation data.

[0075] Step 4.3: Use error propagation analysis and uncertainty assessment to evaluate and analyze the errors in the observation data, and allocate the errors according to their sources and nature.

[0076] Step 5: Using the background forecast field obtained in Step 3, combined with the observation data in Step 4, update the ensemble perturbation field using the ETKF transformation.

[0077] Preferred but not limited, the covariance inflation coefficient is estimated using the average observation information covariance (WG03 scheme), 30 ensemble perturbations are generated using 3 times the grid spacing of deterministic analysis, and the observation spatial positioning scale is set to 400 km.

[0078] Step 6: Based on the ensemble forecast results obtained in Step 3, calculate the ensemble mean and ensemble covariance.

[0079] Step 7: Use En3DVar update ensemble averaging with large-scale analysis constraints, and combine it with observations to generate a hybrid analysis field.

[0080] In a preferred but non-limiting embodiment of the invention, the vertical model layer index for initiating large-scale analysis constraints is 10. To avoid potential error correlations in large-scale GFS information, data refinement is performed at 60 km, approximately twice the GFS resolution. The En3DVar module uses 75% ensemble covariance and 25% static covariance, with a localization scale of 150 km. The static background error covariance matrix is ​​calculated using the NMC (National Meteorological Center) method, with samples consisting of 12-hour and 24-hour forecast fields simulated by WRF for one consecutive month starting from 00:00 UTC on July 1, 2015. Control variables include u, v, temperature, relative humidity, and surface pressure.

[0081] In step 7, large-scale analysis constraints are introduced into the cost function of En3DVar, and the hybrid analysis field is obtained by solving the cost function.

[0082] Based on the idea of ​​extended variables combined with the incremental method, the cost function of the general En3DVar method is expressed by the following formula (3):

[0083]

[0084] In the formula:

[0085] J1 represents the background item related to 3DVar;

[0086] J e Indicates background items related to the set;

[0087] J o Indicates the observation term;

[0088] β1 and β2 represent the weights of the static background error covariance and the ensemble covariance. To ensure the conservation of the total background error variance, β1 and β2 must satisfy the following formula (4):

[0089]

[0090] α represents k vectors α k The matrix formed, i.e. The extended control variables are constrained by a diagonal matrix A, where each matrix block is used to constrain α. k The correlation matrix of spatial variation is expressed by the following formula (5):

[0091]

[0092] A represents the spatial covariance of α;

[0093] B represents the background field error covariance matrix;

[0094] R represents the observation error covariance matrix;

[0095] H represents the linearized observation operator that transforms the state variables of the model space analysis into the state variables of the observation space;

[0096] y o′ This represents the update vector, i.e., y. o′ =y o -H(x b ), y o Let x represent the observation information vector. b It is a background field forecast;

[0097] x′ represents the analytical increment in the data assimilation system after introducing the ensemble forecast covariance matrix of “flow correlation”, expressed by the following formulas (6) and (7):

[0098]

[0099] x1′ represents the analysis increment associated with the static background error covariance of the 3DVar system;

[0100] The second term represents the increment of the set covariance correlation analysis related to "flow correlation", which is a local linear combination of set perturbations;

[0101] x k This represents the forecast for the k-th set. The ensemble forecast average is represented by K, and the number of members in the ensemble forecast system is represented by K.

[0102] α k The extended control variable associated with each set member is a quantity that determines the localized scale of the set covariance and varies with spatial distribution.

[0103] To avoid inverting very large matrices, a series of transformations are performed on the aforementioned cost function, including: using a pre-adjustment transformation technique to decompose the symmetric positive semidefinite real matrix into Uu... T Form; Introducing a new control variable v = U -1 x′, w=(w1,...,w m ) T To eliminate the correlation between the state variables in the analysis, the physical operator U is used. p Transform them into uncorrelated analytical variables. Accordingly, the cost function expression is transformed into the following formula (8):

[0104]

[0105] The expression for calculating x′ is as follows (9):

[0106]

[0107] In the formula:

[0108] U p Represents a physical operator;

[0109] U v Represents the vertical mode mapping operator;

[0110] U h Indicates the horizontal filtering operator;

[0111] Represents the localization operator in the vertical direction;

[0112] Represents the localization operator in the horizontal direction;

[0113] Indicates the predicted perturbation value of the set members;

[0114] w τ This indicates a newly defined control variable. Where α τ Let τ represent the weighting coefficient of the τ-th predicted perturbation field.

[0115] Therefore, the calculation of the cost function and its gradient is transformed into an iterative solution of the new control variable (v, w).

[0116] In this method, large-scale analysis constraints are introduced into the assimilation of mixed En3DVar data through a constraint term, namely the last term of the cost function of the En3DVar method, as expressed by the following formula (10):

[0117]

[0118] In the formula:

[0119] This represents meridional wind u, zonal wind v, temperature t, and water vapor mixing ratio q from global model analysis or forecasts, which are assimilated into large-scale observational data.

[0120] H c This represents a linear operator that maps variables from the grid space of a region model to their locations in a large-scale field.

[0121] R c The large-scale analysis error covariance matrix represents u, v, t, and q, with each variable using a constant, uncorrelated error when constructing the matrix;

[0122] This indicates a large-scale background field prediction, and the subscript 'c' here represents a constraint term.

[0123] For the cost function of the En3DVar data assimilation method that incorporates large-scale analysis constraints, the specific implementation steps include:

[0124] Step 7.1: Read the background field grid data x b And related background error covariance matrix B statistical data;

[0125] Step 7.2: Read the predicted perturbation values ​​of m set members

[0126] In this embodiment, the number of set members is set to 50.

[0127] Step 7.3: Read various types of observation data, including satellite observations. o Large-scale observation data And set the corresponding observation error;

[0128] In a preferred but non-limiting embodiment of the invention, each variable uses a constant uncorrelated error, with meridional and zonal winds at 2.5 m / s, temperature at 2°C, and water vapor mixing ratio at 3 g / kg.

[0129] Step 7.4: Calculate the innovation vector y o -H(x b ),

[0130] Step 7.5: Set the initial values ​​for the analysis control variable (v, w) during iteration, where w is composed of m vector fields w1, w2, ..., w m composition;

[0131] Step 7.6: Calculate the cost function value based on the control variables (v,w), and calculate the background term value, analysis increment value x′, observation term value, and large-scale analysis constraint term value in the cost function in sequence;

[0132] Step 7.7: Calculate the gradient of the cost function with respect to the analysis control variables (v, w), expressed as follows: (11) and (12):

[0133]

[0134] In the formula:

[0135] This represents the gradient value of v.

[0136] This represents the gradient value of w.

[0137] express Matrix operator transpose.

[0138] Step 7.8: Based on the obtained gradient values, calculate the gradient descent direction and optimal step size, and update the iterative values ​​of the control variables (v, w). Then repeat the calculation of the cost function and the gradient value of the cost function (steps 6 and 7) until the convergence criterion is reached;

[0139] Step 7.9: Calculate the final analysis value x a =x b +x′.

[0140] Step 8: Superimpose the perturbation field updated in Step 5 onto the hybrid analysis field generated in Step 7 to form a new set of members, which serves as the new initial field.

[0141] Step 9: Generate a new initial perturbation field and enter the next cycle of ensemble forecasting. Repeat steps 4 to 9 until the forecast period ends.

[0142] It is worth noting that, as a prominent substantive feature and one of the significant advancements achieved by this invention, the generation of more accurate wind field simulations and prediction fields through the above steps can provide high-precision meteorological data support for business scenarios such as wind farm power prediction, offshore wind energy resource assessment, and power meteorological disaster prediction and early warning. In subsequent applications, this invention can also guide power system dispatching and energy allocation, achieving efficient utilization and allocation of wind energy through the rational use of meteorological wind field data, improving the flexibility and stability of the power system; optimizing wind energy resource development and utilization strategies, realizing renewable energy planning and layout based on the analysis of wind and solar resources in different times and spaces, promoting diversified energy utilization and the development of energy storage technologies; and preventing and responding to power system disasters by analyzing meteorological data to predict extreme weather events such as strong winds in advance and take timely measures to ensure the safe operation of power equipment.

[0143] Embodiment 2 of the present invention provides a multi-scale prediction system for new energy wind farms based on large-scale analysis constraints, which runs a multi-scale prediction method for new energy wind farms based on large-scale analysis constraints as described in Embodiment 1, including: a WRF platform and a data acquisition module.

[0144] The data acquisition module is used to acquire conventional observation data and satellite observation data, and input them into the WRF platform for assimilation;

[0145] The WRF platform incorporates large-scale analysis constraints into the cost function of En3Dvar to generate a hybrid analysis field for iterative iteration until the forecast period ends.

[0146] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the multi-scale prediction method for new energy wind farms based on large-scale analysis constraints as described in Embodiment 1.

[0147] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the multi-scale prediction method for new energy wind farms based on large-scale analysis constraints as described in Embodiment 1.

[0148] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0149] Compared with existing technologies, the beneficial technical effects of this invention include at least the following: This invention proposes a multi-scale prediction method for new energy wind farms based on large-scale analysis constraints, including: WRF configuration and physical process parameterization, establishing initial and boundary fields, calculating background forecast fields, observation data assimilation, updating ensemble perturbation fields, calculating ensemble averages and ensemble covariance, generating a mixed analysis field using a mixed data assimilation system, generating a new initial field and a new initial perturbation field, and iteratively forecasting until the forecast is completed. This method uses the En3DVar data assimilation framework to assimilate meteorological satellite observation data and adds global model information as an additional observation value to the variational cost function, thereby minimizing the large-scale error in the initial dynamic field analysis. Compared with existing prediction capabilities, it generates more accurate simulation and prediction fields for wind fields, especially deep-sea wind fields, providing high-precision meteorological data support for operational scenarios such as wind farm power prediction, offshore wind energy resource assessment, and power meteorological disaster prediction and early warning, thereby improving the accuracy of power prediction, wind energy resource assessment, and power meteorological disaster prediction.

[0150] More application scenarios include, but are not limited to: guiding power system dispatch and energy allocation, achieving efficient utilization and allocation of wind energy through the rational use of meteorological wind field data, and improving the flexibility and stability of the power system; optimizing wind energy resource development and utilization strategies, realizing renewable energy planning and layout based on the analysis of wind and solar resources in different times and spaces, and promoting the development of diversified energy utilization and energy storage technologies; preventing and responding to power system disasters, predicting extreme weather events such as strong winds in advance by analyzing meteorological data and taking timely measures to ensure the safe operation of power equipment.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A multi-scale prediction method for new energy wind farms based on large-scale analysis constraints, characterized in that, Includes the following steps: Configure the mesoscale weather numerical forecasting model WRF model and parameterize the physical processes; use real-time forecast data from NCEP, interpolate the initial field through the WPS initialization module to obtain the initial field, and use GFS data to establish the lateral boundary conditions; Starting from the NCEP analysis values, integrate over several hours to obtain several hours of RMSE perturbation profiles, generate K random perturbations, perform several hours of ensemble forecasting, and obtain a background forecast field of K members; Read conventional observation data and satellite wind observation data, and assimilate them; use K-member background forecast field combined with observation data, and use ETKF transformation to update the ensemble perturbation field; Based on the ensemble forecast results obtained from several hours of ensemble forecasts, the ensemble mean and ensemble covariance are calculated; large-scale analysis constraints are introduced into the cost function of En3Dvar, the ensemble mean is updated based on the ensemble mean and ensemble covariance, and combined with observation data, a mixed analysis field is generated. The updated ensemble perturbation field is superimposed onto the hybrid analysis field to form a new set of ensemble members, which serves as the new initial field. A new initial perturbation field is generated, and the ensemble forecast enters the next cycle. The process is iteratively executed until the forecast period ends.

2. The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints as described in claim 1, characterized in that: The configured mesoscale weather numerical forecasting model WRF model includes: Select the forecast model and data assimilation system; and use a two-layer nested grid with a horizontal grid spacing of 5 km, 45 vertical layers, and a 50 hPa model top; The parameterization of the physical process includes: Radiation processes, microphysical processes, boundary layer schemes, cumulus convection parameterization, and road surface processes were studied using the WSM6 microphysical scheme, the RRTMGT radiative transfer scheme, the Tiedtke cumulus convection scheme, the YSU planetary boundary layer scheme, the modified MM5 surface layer scheme, and the Noah land surface model scheme.

3. The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to claim 1, characterized in that: The conventional observation data are from GDAS provided by NCEP, including: SYNOP, METAR, SHIP+BUOY, GTS radiosonde observation data, AMDAR, and SATWND; The satellite-observed winds include: cloud-guided winds, sea surface winds from China's Wind Catcher-1 A / B satellites, and atmospheric motion vectors based on radiance from Japan's Himawari-8 meteorological satellite.

4. The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to claim 3, characterized in that: The assimilation of conventional observation data and satellite observation data includes: Wind observation data from different sources and in different formats are uniformly organized into a format suitable for assimilation processing; Denoising, missing value imputation, and / or data smoothing algorithms are used to control the quality of observational data. Error propagation analysis and uncertainty assessment are used to evaluate and analyze the errors in the observation data, and the errors are allocated according to their sources and nature.

5. A multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to any one of claims 1 to 4, characterized in that: Using NCEP's GFS data, interpolation was performed through the WPS initialization module, and the initial field was obtained during continuous loop En3DVar operation. The data was assimilated every 12 hours, and the lateral boundary conditions were provided by GFS reanalysis data every 6 hours. Integrating the NCEP analysis values ​​over 6 hours yields the 6-hour RMSE perturbation profile, generating 30 random perturbations. A 6-hour ensemble forecast is then performed to obtain the background forecast field for 30 members.

6. A multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to any one of claims 1 to 4, characterized in that: The cost function that incorporates large-scale analysis constraints into En3Dvar includes: A pre-adjustment transformation is used to transform the original cost function of the En3DVar method, converting the calculation of the original cost function and its gradient into an iterative solution of the new control variables; Large-scale analysis constraints are added to the transformed cost function to incorporate them into the assimilation of hybrid En3DVar data.

7. The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to claim 6, characterized in that: The large-scale analysis constraint term is expressed by the following formula (1): In the formula: This refers to meteorological data derived from global model analysis or forecasts and assimilated into large-scale observational data. H c This represents a linear operator that maps variables from the grid space of a region model to their locations in a large-scale field. R c This represents the error covariance matrix of large-scale analysis of meteorological data. x ′ This represents the analytical increment in the data assimilation system after introducing the ensemble forecast covariance matrix of "flow correlation". This indicates a large-scale background field forecast.

8. The multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to claim 7, characterized in that: The process of updating the ensemble mean based on ensemble mean and ensemble covariance, and generating a mixed analysis field by combining observational data, includes: The introduction of large-scale analysis constraints into the assimilation of hybrid En3DVar data includes: Read background field grid data x b And related background error covariance matrix B statistical data; Read the predicted perturbation values ​​of m set members Reading various types of observation data, including satellite observations. o Large-scale observation data And set the corresponding observation error; Calculate the innovation vector y in the cost function o -H(x b ), Set the initial values ​​for the analysis control variable (v, w) during iteration, where w is composed of m vector fields w1, w2, ..., w m composition; Calculate the cost function value based on the control variables (v, w), and then calculate the background term value and the analysis increment value x in the cost function in sequence. ′ Observational values, large-scale analysis constraint values; Calculate the gradient of the cost function with respect to the control variables (v, w); Based on the obtained gradient values, calculate the gradient descent direction and optimal step size, and update the iterative values ​​of the control variables (v, w) until the convergence criterion is reached; The final analytical value is obtained by incremental calculation of background field forecast overlay analysis.

9. A multi-scale prediction method for new energy wind farms based on large-scale analysis constraints as described in claim 7 or 8, characterized in that: The vertical model layer index for initiating large-scale analysis constraints is set to 10, and data refinement is performed over a distance of 60 kilometers. The En3DVar module uses 75% ensemble covariance and 25% static covariance, with a localization scale of 150km. The static background error covariance matrix is ​​calculated using the NMC method.

10. A multi-scale prediction system for new energy wind farms based on large-scale analysis constraints, running the multi-scale prediction method for new energy wind farms based on large-scale analysis constraints according to any one of claims 1 to 9, characterized in that, include: WRF platform, data acquisition module: The data acquisition module is used to acquire conventional observation data and satellite observation data, and input them into the WRF platform for assimilation; The WRF platform incorporates large-scale analysis constraints into the cost function of En3Dvar to generate a hybrid analysis field for iterative iteration until the forecast period ends.

Images