An automatic optimization wind resource prediction method based on a multi-parameterization scheme optimization and WRF nesting optimization

Abstract

This invention discloses an automatic optimization method for wind resource forecasting based on multi-parameterized scheme selection and WRF nested optimization. The method includes: automatically acquiring and standardizing GFS meteorological, geographic static, and high-precision topographic data; constructing a fixed parameterized scheme library, which can randomly generate extended schemes as needed; configuring key WRF physical processes according to scenario adaptability principles; dynamically determining WRF nested schemes and densifying the vertical layer near the hub height; running each scheme through multi-instance parallel computation to obtain simulated data of key meteorological elements; constructing a comprehensive evaluation system based on multiple indicators for quantitative scoring and dynamically adjusting scenario adaptability principles; selecting the optimal scheme, restarting the WRF simulation, and outputting high-resolution wind field data. This invention achieves intelligent selection and nested optimization of parameterized schemes, improving the accuracy and automation level of wind resource forecasting, and is adaptable to different scenario requirements.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to an automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization. Background Technology

[0002] Wind energy, as a clean and renewable energy source, plays a crucial strategic role in global efforts to address climate change, reduce greenhouse gas emissions, and decrease dependence on fossil fuels, and is particularly vital for achieving my country's carbon peaking and carbon neutrality goals. However, wind resources are characterized by significant randomness and intermittency, which not only increases the uncertainty of wind power grid connection but also places higher demands on the stability and absorption capacity of the power system. Therefore, conducting high-precision wind speed forecasting has become a core requirement for improving wind farm operating efficiency and ensuring grid security.

[0003] Existing wind speed forecasting methods are mostly based on statistical models derived from historical observation data. These methods rely on multi-source observation data from ground-based meteorological stations, wind towers, radiosondes, radar, and satellite remote sensing, achieving a certain level of accuracy in ultra-short-term forecasts (within hours). However, as the forecast lead time increases, errors accumulate rapidly, leading to a significant decrease in forecast accuracy. Furthermore, the widely adopted near-surface wind speed assessment theories are typically based on similarity theories assuming a constant flux-gradient relationship, and these theories are generally applicable at altitudes not exceeding 100m. However, with the increasing size and height of wind turbines, the hub height of these turbines generally exceeds the traditional near-surface application range, resulting in significant limitations for similarity theory-based methods in upper-altitude wind speed forecasting, making it difficult to meet the needs of engineering applications.

[0004] To address the shortcomings of traditional methods, mesoscale numerical weather prediction (WRF) models have gradually become an important tool for wind speed forecasting. WRF models can characterize complex atmospheric physical processes, support multi-scale nesting and data assimilation, and are suitable for short- to medium-term wind speed simulation. However, the forecasting performance of WRF models is highly dependent on the parameterization scheme of the physical processes. Different parameterization schemes directly affect the simulation accuracy of energy transport and dynamic mechanisms by characterizing processes such as cloud microphysics, cumulus convection, boundary layer turbulence, and radiative transfer. Existing research shows that different parameterization schemes exhibit significant differences in simulation results for different regions and meteorological elements, and even show different applicability under different topographic conditions within the same region. This fully demonstrates that the choice of parameterization scheme has a decisive impact on forecast accuracy.

[0005] Although existing technologies have recognized the importance of screening parameterized schemes for WRF and developed various approaches, such as fusing results from multiple schemes through post-processing artificial intelligence models, determining static optimal configurations offline for specific regions, or assimilating and correcting the initial field using observational data (for example, CN120805704A discloses a cross-scale wind farm wind resource assessment method based on numerical models, and CN120525133B discloses a high-precision correction method and system for wind speed forecasting for power system wind power prediction), these methods still have fundamental limitations when facing the challenge of universally applicable screening of massive combinations of parameterized schemes: their screening scope is often limited to preset, finite combinations of known schemes, lacking the ability to systematically explore a broader scheme space, thus making it difficult to guarantee that the selected scheme is globally or situationally optimal. Summary of the Invention

[0006] To overcome the limitations of existing technologies, such as limited exploration space for parameterized scheme combinations, low screening efficiency, and single evaluation dimensions, this invention proposes an automatic optimization wind resource forecasting method based on multi-parameterized scheme selection and WRF nested optimization. By constructing a pre-set parameterized scheme library and combining it with a scheme expansion mechanism, and employing a comprehensive evaluation system covering multiple meteorological elements such as temperature, air pressure, wind speed, and wind direction, this method automatically compares and optimizes key physical process scheme combinations. It achieves refined simulation of wind fields under different terrain and atmospheric conditions, and is more scientific and comprehensive than traditional single-dimensional evaluations relying solely on wind speed. Through this multi-dimensional quantitative scoring mechanism, this invention can automatically and dynamically select the optimal parameterized scheme combination best suited to specific meteorological and geographical conditions. This not only fundamentally improves the adaptability and accuracy of numerical forecasting but also provides reliable and comprehensive decision support for key engineering applications such as refined wind farm site selection, turbine layout optimization, and wake planning.

[0007] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0008] An automatic optimization method for wind resource forecasting based on multi-parameter scheme selection and WRF nested optimization, the method comprising:

[0009] S1 automatically acquires GFS meteorological driving data, geographic static data and high-precision terrain data of the target wind farm, performs integrity verification and format standardization on the acquired data, and obtains input field data for subsequent simulation.

[0010] S2, construct a fixed parameterization scheme library containing a set of typical physical parameterization combination schemes, and select whether to randomly generate parameterization extension schemes based on the underlying surface and meteorological conditions data of the target wind farm or the control commands input by the user. If so, according to the requirements of wind farm engineering application for the simulation accuracy and stability of key meteorological elements, and in accordance with the scenario adaptability preset principle, automatically parameterize the key physical processes of WRF mode to generate multiple extended parameterization schemes, and proceed to step S3; otherwise, directly proceed to step S3.

[0011] S3. Based on the location of the forecast points, dynamically determine the WRF nesting scheme, which includes the number of nesting layers, the grid resolution and range of each layer; execute the WPS module in sequence to generate the initial field and boundary conditions of the model; dynamically adjust the vertical layer height of the eta coordinate to make the model layer match the hub height of the wind turbine, and refine the vertical layer near the corresponding hub height.

[0012] S4 employs a multi-instance WRF parallel computing framework to simultaneously run fixed parameterization schemes and extended parameterization schemes, acquiring simulation data of key meteorological elements, including wind speed, temperature, air pressure, and wind direction, under each scheme.

[0013] S5 performs spatiotemporal matching and processing of key meteorological element simulation data and corresponding observation data. Based on mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and multiple meteorological indicators including wind direction frequency distribution, a comprehensive evaluation system is constructed to quantitatively score the simulation performance of each parameterized scheme and adjust the scenario adaptability preset principle according to the quantitative score.

[0014] S6 quantifies and ranks the simulation performance of each parameterization scheme, automatically selects the parameterization scheme with the best overall performance based on the ranking results, and restarts the WRF mode with this parameterization scheme to perform the final forecast simulation, outputting hub height and multi-level high-resolution wind field data.

[0015] Step S1 further includes:

[0016] Determine the latitude and longitude of the prediction point and the target prediction time period for the numerical simulation of the target wind farm; based on the set target prediction time period, set the start and end times of the meteorological driving field, and support automatic downloading at four times: UTC00, 06, 12, and 18.

[0017] Download SRTM3s resolution elevation data tiles near the prediction point, merge the hgt format data and convert it to tif format, and then further convert it to geogrid binary format; then create a folder named srtm_3s under the geog_data_path path in the namelist.wps configuration file, and place the processed elevation data and index files in this folder directory for subsequent modeling and simulation.

[0018] Furthermore, in step S2, based on the requirements of wind farm engineering applications for the simulation accuracy and stability of key meteorological elements, and following the principle of scenario adaptability preset, the process of automatically parameterizing the key physical processes of the WRF mode and generating multiple extended parameterization schemes includes:

[0019] The simulation accuracy and stability requirements of key meteorological elements for wind farm engineering applications, including hub height wind speed error and wind direction prediction deviation, are quantitatively analyzed and matched with the key physical processes of the WRF model. The key physical processes of the WRF model include microphysical processes, long-wave and short-wave radiation processes, boundary layer turbulent mixing mechanism, near-surface layer flux exchange process, cumulus convection process and land surface process.

[0020] Based on the matching results, and in accordance with the principle of scenario adaptability, one or more candidate schemes for each key physical process are selected from the preset scheme parameter library.

[0021] Based on the parameter combination of the fixed parameterization scheme, and with the core constraint of maintaining the forced coordination between the boundary layer and near-ground layer schemes, multiple sets of extended parameterization schemes with differentiated configurations are generated by replacing single physical process schemes or replacing single sets of coordinated physical process schemes. Specifically, for key physical processes that are not coordinated, the single physical process scheme replacement method is adopted, and for physical process schemes with coordinated logic, the single set of coordinated physical process scheme replacement method is adopted.

[0022] Step S3 further includes:

[0023] Based on the latitude and longitude coordinates of the target wind farm, a Lambert conformal conic projection method is adopted, and a triple-nested grid system is constructed according to a horizontal nesting ratio of 1:3:3. The outermost grid covers the mesoscale region where the target wind farm is located, and is used to introduce the large-scale circulation background. The middle layer covers the regional scale where the target wind farm is located, and is used to achieve a smooth transition from large-scale to small-scale meteorological fields. The innermost high-resolution core domain directly covers the target wind farm area to support the fine simulation of the wind farm.

[0024] In the vertical direction, the terrain-following eta coordinate system is adopted, and the total number of vertical layers in the whole area is set. The vertical layer densification is carried out near the height of the wind turbine hub through an adaptive optimization algorithm. The layer spacing of the layers above and below the hub height is smaller than that of other layers, so that the hub height area has higher vertical resolution to characterize the wind speed profile, turbulence structure and boundary layer thermodynamic characteristics at the hub height.

[0025] Step S5 further includes:

[0026] S51, Calculate the wind speed sub-score for a single height level. :

[0027] ;

[0028] Then, the average value is taken for all height layers containing wind speed observation data to obtain the comprehensive score of wind speed element. :

[0029] ;

[0030] in, The total number of height layers containing wind speed observation data, where h is the index of a single height layer. , and Let represent the normalized scores of the mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) of the wind speed at the h-th altitude layer for the i-th physical parameterization scheme, respectively.

[0031] S52, the wind direction records during the observation period are statistically analyzed according to 16 sectors, and the average absolute angular error of the i-th scheme is... :

[0032] ;

[0033] Where K=16 represents the total number of wind direction sectors, and k is the sector index. Let be the average wind direction of the i-th scheme in the k-th sector. The average wind direction observed in the corresponding sector;

[0034] Calculate the wind direction angle error score :

[0035] ;

[0036] In the formula, and These represent the maximum and minimum absolute errors of the wind direction angle for all schemes;

[0037] S53, after taking the arithmetic mean of the deviations in the frequency of wind direction occurrence in each sector, the wind direction frequency error score is calculated. ;

[0038] S54 uses a weighted average method to calculate the comprehensive score of wind direction element. :

[0039] ;

[0040] In the formula, and These are the angle error weight and the frequency error weight, respectively, with the angle error weight being greater than the frequency error weight.

[0041] S55, Calculate the normalized score of single-level atmospheric pressure error. Normalized score of temperature error at single height level :

[0042] ;

[0043] ;

[0044] In the formula, and Let represent the average absolute error of air pressure and the average absolute error of temperature at the h-th altitude layer for the i-th parameterization scheme, respectively. and These represent the maximum and minimum MAE values ​​of air pressure at the h-th altitude layer for all parameterization schemes; and These represent the maximum and minimum MAE values ​​for temperature at the h-th altitude layer for all parameterization schemes;

[0045] Then, the average value is taken for all height layers containing observation data to obtain the comprehensive score of air pressure elements. Comprehensive score based on temperature factor :

[0046] ;

[0047] ;

[0048] In the formula, The total number of altitude layers containing both air pressure and temperature observation data, where h is the index of a single altitude layer;

[0049] S56 assigns weights to each meteorological element based on its engineering importance, and calculates the final score using a weighted linear overlay method. :

[0050] ;

[0051] In the formula, , , , These represent the comprehensive scores of the i-th parameterization scheme for wind speed, wind direction, temperature, and air pressure, respectively. , , and These represent the weights of wind speed, wind direction, temperature, and air pressure, respectively. Among them, wind speed has the highest weight, followed by wind direction, while temperature and air pressure have the lowest weights.

[0052] Furthermore, in step S5, the process of adjusting the preset principles for scene adaptability based on the quantitative score includes:

[0053] The quantitative scoring results, target wind farm scene features, and corresponding parameterization schemes are combined and stored together to form a structured dataset; scene features, including terrain type, meteorological conditions, and hub height, are quantified according to preset categories.

[0054] For the same type of scenario, the top-scoring parameterized schemes are selected, and the common characteristics of their key physical process schemes are extracted to identify the optimal scheme combination type for that scenario. Then, the score changes of the same parameterized scheme in different scenarios are analyzed to obtain the scenario adaptation preference of the parameterized scheme. Finally, for each scenario, the contribution ratio of the comprehensive score of wind speed, wind direction, temperature, and air pressure to the final score is calculated to determine the key meteorological factors affecting the suitability of the scheme in each scenario, and the mapping relationship between the score and the suitability of the parameterized scheme is analyzed.

[0055] Based on the mapping relationship between scores and the adaptability of parameterized schemes, the priority of parameterized scheme selection is adjusted, and the physical process schemes corresponding to the common characteristics of high-scoring schemes are listed as the priority selection options for the corresponding scenarios; at the same time, scenario-specific adaptation rules are added to clarify the optimal scheme combination type of key physical processes under different scenarios; and the selection weight of corresponding physical process schemes is adjusted according to the contribution ratio of meteorological elements.

[0056] Apply the adjusted scenario adaptability preset principle to the new simulation task and compare the changes in the overall score before and after the update. If the overall score of the optimal solution improves under the new principle and the score differentiation of different solutions improves, then the update is confirmed to be effective; otherwise, re-analyze the correlation between the quantitative score and the parameterized solution.

[0057] Furthermore, in step S5, the timing of adjusting the scene adaptability preset principle is controlled by setting update trigger conditions.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] First, the automatic optimization wind resource forecasting method based on multi-parameterized scheme selection and WRF nested optimization of the present invention addresses the problems of traditional methods, such as strong reliance on human experience, insufficient scheme space coverage, and evaluation indicators being limited to a single wind speed factor when selecting the optimal parameterized scheme. The present invention deeply embeds the intelligent optimization mechanism into the entire forecasting process. By constructing a fixed scheme library containing a variety of classic configurations as an efficient screening starting point, and having the ability to dynamically generate and test diverse new parameter combinations, it realizes a systematic and proactive exploration from known schemes to unknown spaces.

[0060] Secondly, the automatic optimization wind resource forecasting method of this invention, based on multi-parameter scheme selection and WRF nested optimization, adopts a comprehensive evaluation system integrating multiple meteorological elements such as temperature, air pressure, wind speed, and wind direction. This system enables a scientific and comprehensive quantitative evaluation of the simulation performance of parameterized schemes. Through this closed-loop optimization architecture, the optimal combination of parameterized schemes best suited to the meteorological and geographical characteristics of the target area can be automatically and dynamically selected. This significantly improves the objectivity, efficiency, and engineering applicability of scheme selection, providing a more reliable and refined wind resource data foundation for wind farm site selection, turbine layout optimization, and wake analysis. Attached Figure Description

[0061] Figure 1 The flowchart of the automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization of the present invention is shown below.

[0062] Figure 2a This is a schematic diagram of the elevation distribution of wind fields in complex hilly areas;

[0063] Figure 2b This is a schematic diagram of the elevation distribution of wind fields in the plains;

[0064] Figure 3a It is a graph showing the daily average temperature variation curves of the near-surface layer under different combinations of parameterization schemes for a plain wind field;

[0065] Figure 3b It is a graph showing the daily average variation of near-surface air pressure under different parameterization schemes for a plain wind field;

[0066] Figure 3c It is a graph showing the daily average wind speed variation at a height of 70m for different combinations of parameterization schemes in a plain wind field;

[0067] Figure 3d This is a graph showing the daily average wind speed variation at a height of 120m for different combinations of parameterization schemes in a plain wind field. Detailed Implementation

[0068] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0069] This invention proposes an automatic optimization method for wind resource forecasting based on multi-parameter scheme selection and WRF nested optimization, characterized in that the method includes:

[0070] S1 automatically acquires GFS meteorological driving data, geographic static data and high-precision terrain data of the target wind farm, performs integrity verification and format standardization on the acquired data, and obtains input field data for subsequent simulation.

[0071] S2, construct a fixed parameterization scheme library containing a set of typical physical parameterization combination schemes, and select whether to randomly generate parameterization extension schemes based on the underlying surface and meteorological conditions data of the target wind farm or the control commands input by the user. If so, according to the requirements of wind farm engineering application for the simulation accuracy and stability of key meteorological elements, and in accordance with the scenario adaptability preset principle, automatically parameterize the key physical processes of WRF mode to generate multiple extended parameterization schemes, and proceed to step S3; otherwise, directly proceed to step S3.

[0072] S3. Based on the location of the forecast points, dynamically determine the WRF nesting scheme, which includes the number of nesting layers, the grid resolution and range of each layer; execute the WPS module in sequence to generate the initial field and boundary conditions of the model; dynamically adjust the vertical layer height of the eta coordinate to make the model layer match the hub height of the wind turbine, and refine the vertical layer near the corresponding hub height.

[0073] S4 employs a multi-instance WRF parallel computing framework to simultaneously run fixed parameterization schemes and extended parameterization schemes, acquiring simulation data of key meteorological elements, including wind speed, temperature, air pressure, and wind direction, under each scheme.

[0074] S5 performs spatiotemporal matching and processing of key meteorological element simulation data and corresponding observation data. Based on mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and multiple meteorological indicators including wind direction frequency distribution, a comprehensive evaluation system is constructed to quantitatively score the simulation performance of each parameterized scheme and adjust the scenario adaptability preset principle according to the quantitative score.

[0075] S6 quantifies and ranks the simulation performance of each parameterization scheme, automatically selects the parameterization scheme with the best overall performance based on the ranking results, and restarts the WRF mode with this parameterization scheme to perform the final forecast simulation, outputting hub height and multi-level high-resolution wind field data.

[0076] Step S1 further includes:

[0077] GFS meteorological driving data acquisition and processing: Automatically download GFS meteorological driving data from the National Center for Atmospheric Research (NCAR). Specifically, determine the latitude and longitude of the prediction points and the target prediction time period for the numerical simulation. Based on the set prediction time range, set the start and end times of the meteorological driving field, and support automatic download at four time points: UTC00, 06, 12, and 18.

[0078] The acquisition and processing of static terrain data includes: downloading SRTM3s resolution elevation data tiles near the prediction point, merging the hgt format data and converting it to tif format, and then further converting it to geogrid binary format; then creating a folder named srtm_3s under the geog_data_path path in the namelist.wps configuration file, and placing the processed elevation data and index files in this directory for subsequent modeling and simulation.

[0079] Step S2 further includes:

[0080] This paper quantifies and analyzes the simulation accuracy and stability requirements of key meteorological elements in wind farm engineering applications, including hub height wind speed error and wind direction prediction deviation, and matches them with the key physical processes of the WRF model. These key physical processes include microphysical processes, longwave and shortwave radiation processes, boundary layer turbulent mixing mechanisms, near-surface flux exchange processes, cumulus convection processes, and land surface processes. For example, when focusing on wind speed accuracy requirements under high wind speeds, the focus is on matching boundary layer turbulent mixing mechanisms and near-surface flux exchange processes; when focusing on wind direction stability requirements under complex terrain, the focus is on matching land surface processes and boundary layer turbulent mixing mechanisms; and when focusing on meteorological element accuracy requirements in precipitation / cloudy scenarios, the focus is on matching microphysical processes and longwave / shortwave radiation processes.

[0081] Based on the matching results, and following the pre-defined principles of scenario adaptability, one or more candidate schemes for each key physical process are selected from the pre-defined scheme parameter library. Preferably, the pre-defined principles of scenario adaptability can be set from three aspects: scenario matching degree, historical performance of simulation accuracy, and computational efficiency. Among these, scenario matching degree is a hard requirement, while historical performance of simulation accuracy and computational efficiency are optional. One or more candidate schemes for each key physical process are selected from the pre-defined scheme parameter library. Below are some candidate schemes corresponding to key physical processes: WSM6 (for warm cloud scenarios) and Thompson (for cold / hybrid cloud scenarios) for microphysical processes, long wave and The candidate scenarios are: RRTM-Dudhia (for mid-to-low latitude scenarios) and RRTMG-RRTMG (for high-latitude strong radiation scenarios) for shortwave radiation processes; YSU (for neutral / unstable stratification) and ACM2 (for stable stratification) for boundary layer turbulent mixing mechanisms; MM5 (for flat underlying surfaces) and Eta (for complex underlying surfaces) for near-surface flux exchange processes; Kain-Fritsch (for outer-layer nested mesoscale convection) and closed configuration (for inner high-resolution regions) for cumulus convection processes; and Noah (for vegetated areas) and Noah-MP (for arid / semi-arid regions) for land surface processes. These candidate scenarios are for illustrative purposes only and do not imply that each key physical process has only these few candidate scenarios.

[0082] Based on the parameter combinations of fixed parameterization schemes, and with the core constraint of maintaining the forced coordination between boundary layer and near-surface layer schemes, such as YSU and MM5, and ACM2 and Eta being fixed adaptation combinations, multiple sets of extended parameterization schemes with differentiated configurations are generated by replacing single physical process schemes or single sets of coordinated physical process schemes. Among them, for key physical processes that are not coordinated, such as microphysical processes, long-wave and short-wave radiation processes, land surface processes, and cumulus convection processes, the single physical process scheme replacement method is adopted. For physical process schemes with coordinated logic, such as boundary layer turbulent mixing mechanism and near-surface layer flux exchange process, microphysical process and long-wave / short-wave radiation process, and land surface process and cumulus convection process, the single set of coordinated physical process scheme replacement method is adopted.

[0083] Step S3 further includes:

[0084] Based on the latitude and longitude coordinates of the target wind farm, a Lambert conformal conic projection method is adopted, and a triple-nested grid system is constructed according to a 1:3:3 horizontal nesting ratio, i.e., the resolution of the middle layer grid is 1 / 3 of that of the outermost layer, and the resolution of the innermost layer is 1 / 3 of that of the middle layer. The outermost layer grid covers the mesoscale region where the target wind farm is located, used to introduce the large-scale circulation background. The middle layer covers the regional scale where the target wind farm is located, used to achieve a smooth transition from large-scale to small-scale meteorological fields. The innermost high-resolution core domain directly covers the target wind farm area to support refined wind farm simulation. For example, the resolution of the outermost layer grid is set to 15~30km, covering the mesoscale region where the target wind farm is located, used to introduce the large-scale circulation background field; the resolution of the middle layer grid is set to 5~10km, covering the regional scale where the target wind farm is located, used to achieve a smooth transition from large-scale to small-scale meteorological fields; and the resolution of the innermost high-resolution core domain grid is set to 1~3km, accurately covering the entire target wind farm area and a surrounding 5~10km range, to support refined wind farm simulation.

[0085] In the vertical direction, the terrain-following eta coordinate system is adopted, and the total number of vertical layers in the whole area is set. The vertical layer densification is carried out near the hub height of the wind turbine (such as within 300 meters, which is adaptively adjusted according to the actual hub height of the wind farm) through an adaptive optimization algorithm. The layer spacing of the layers above and below the hub height is smaller than that of other layers, so that the hub height area has higher vertical resolution. This configuration can accurately depict the wind speed profile, turbulence intensity distribution and boundary layer thermodynamic characteristics at the hub height, providing high-precision vertical data support for wind resource prediction.

[0086] Step S4 further includes:

[0087] Based on the physical scheme combination configured in step S3, the system launches a multi-instance WRF parallel computing framework to synchronously drive different parameterized schemes to conduct numerical simulation experiments. The system automatically schedules computing resources to achieve task-level parallelism and efficient computation. After the simulation is completed, it automatically extracts key meteorological element output data such as wind speed, wind direction, temperature, and air pressure at different altitude levels (including hub height) for each scheme, providing a standardized data foundation for subsequent comprehensive performance evaluation.

[0088] Step S5 further includes:

[0089] S51, Calculate the wind speed sub-score for a single height level. :

[0090] ;

[0091] Then, the average value is taken for all height layers containing wind speed observation data to obtain the comprehensive score of wind speed element. :

[0092] ;

[0093] in, The total number of height layers containing wind speed observation data, where h is the index of a single height layer. , and Let represent the normalized scores of the mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) of the wind speed at the h-th altitude layer for the i-th physical parameterization scheme, respectively.

[0094] S52, the wind direction records during the observation period are statistically analyzed according to 16 sectors, and the average absolute angular error of the i-th scheme is... :

[0095] ;

[0096] Where K=16 represents the total number of wind direction sectors, and k is the sector index. Let be the average wind direction of the i-th scheme in the k-th sector. The average wind direction observed in the corresponding sector;

[0097] Calculate the wind direction angle error score :

[0098] ;

[0099] In the formula, and These represent the maximum and minimum absolute errors of the wind direction angle for all schemes;

[0100] S53, after taking the arithmetic mean of the deviations in the frequency of wind direction occurrence in each sector, the wind direction frequency error score is calculated. ;

[0101] S54 uses a weighted average method to calculate the comprehensive score of wind direction element. :

[0102] ;

[0103] In the formula, and These are the angle error weight and the frequency error weight, respectively, with the angle error weight being greater than the frequency error weight.

[0104] S55, Calculate the normalized score of single-level atmospheric pressure error. Normalized score of temperature error at single height level :

[0105] ;

[0106] ;

[0107] In the formula, and Let represent the average absolute error of air pressure and the average absolute error of temperature at the h-th altitude layer for the i-th parameterization scheme, respectively. and These represent the maximum and minimum MAE values ​​of air pressure at the h-th altitude layer for all parameterization schemes; and These represent the maximum and minimum MAE values ​​for temperature at the h-th altitude layer for all parameterization schemes;

[0108] Then, the average value is taken for all height layers containing observation data to obtain the comprehensive score of air pressure elements. Comprehensive score based on temperature factor :

[0109] ;

[0110] ;

[0111] In the formula, The total number of altitude layers containing both air pressure and temperature observation data, where h is the index of a single altitude layer;

[0112] S56, weights are assigned according to the engineering importance of each meteorological element, and the final score is calculated using a weighted linear overlay:

[0113] ;

[0114] In the formula, , , , These represent the comprehensive scores of the i-th parameterization scheme for wind speed, wind direction, temperature, and air pressure, respectively. , , and These represent the weights of wind speed, wind direction, temperature, and air pressure, respectively. Among them, wind speed has the highest weight, followed by wind direction, while temperature and air pressure have the lowest weights.

[0115] Step S5, the process of adjusting the preset principles for scene adaptability based on the quantitative score, includes:

[0116] The quantitative scoring results, target wind farm scene features, and corresponding parameterization schemes are combined and stored together to form a structured dataset; scene features, including terrain type, meteorological conditions, and hub height, are quantified according to preset categories.

[0117] For the same type of scenario, the top-scoring parameterized schemes are selected, and the common characteristics of their key physical process schemes are extracted to identify the optimal scheme combination type for that scenario. Then, the score changes of the same parameterized scheme in different scenarios are analyzed to obtain the scenario adaptation preference of the parameterized scheme. Finally, for each scenario, the contribution ratio of the comprehensive score of wind speed, wind direction, temperature, and air pressure to the final score is calculated to determine the key meteorological factors affecting the suitability of the scheme in each scenario, and the mapping relationship between the score and the suitability of the parameterized scheme is analyzed.

[0118] Based on the mapping relationship between scores and the adaptability of parameterized schemes, the priority of parameterized scheme selection is adjusted, and the physical process schemes corresponding to the common characteristics of high-scoring schemes are listed as the priority selection options for the corresponding scenarios; at the same time, scenario-specific adaptation rules are added to clarify the optimal scheme combination type of key physical processes under different scenarios; and the selection weight of corresponding physical process schemes is adjusted according to the contribution ratio of meteorological elements.

[0119] Apply the adjusted scenario adaptability preset principle to the new simulation task and compare the changes in the overall score before and after the update. If the overall score of the optimal solution improves under the new principle and the score differentiation of different solutions improves, then the update is confirmed to be effective; otherwise, re-analyze the correlation between the quantitative score and the parameterized solution.

[0120] To avoid frequent updates, preferably, the timing of adjustments to the preset principles of scene adaptability is controlled by setting update trigger conditions. For example, the principle is updated when the cumulative score of the newly added scene simulation reaches a set number of groups, the fluctuation of the comprehensive score of the optimal solution for a certain type of scene exceeds a preset fluctuation threshold, or the score of the newly added parameterized solution is significantly higher than that of the existing solution.

[0121] Step S6 further includes:

[0122] Based on the ranking results of the comprehensive evaluation system, the system automatically selects the parameterized scheme combination with the best overall performance. Subsequently, based on this optimized scheme, the WRF mode is restarted to perform the final forecast simulation, directly outputting hub height and multi-level high-resolution wind field data, providing automated and high-precision wind resource forecasting products for wind farm site selection, power prediction, and operation optimization.

[0123] Example

[0124] The wind farm areas used for method validation were a hilly area in Dalian, Liaoning Province, and a plain area in Bozhou, Anhui Province. These two locations have complex meteorological conditions and different geographical environments, making them typical cases for studying the wind flow characteristics of complex plains and hilly areas. The former's wind farm topography is mainly hilly and lakey (e.g., Figure 2aAs shown), the anemometer tower has an altitude of 141m, with observation levels at 30m, 50m, 60m, and 70m. The latter wind field topography is mainly plains (e.g., Figure 2b As shown in the figure, the wind measurement tower has an altitude of 32m, and the observation heights are 50m, 80m, 100m, and 120m. The wind resource assessment process in this example strictly follows steps S1 to S6 of this invention, as follows:

[0125] Step 1: Automatically optimize and download hourly forecast data from GFS, and ensure data reliability through integrity verification and timeliness monitoring.

[0126] The system automatically acquires mesoscale meteorological driving data and high-precision topographic data for the target wind farm. Climate input data uses GFS forecast data released by the U.S. National Center for Environmental Prediction, with a horizontal resolution of 0.25° × 0.25°. The WRF simulation period for hilly areas is from 08:00 on May 1, 2017 to 08:00 on May 25, 2017 (Beijing time), and for plain areas, it is from 08:00 on August 1, 2017 to 08:00 on August 30, 2017. Forecast data is input every 6 hours, with an outermost integration step of 27 seconds. Simultaneously, SRTM3s resolution elevation data tiles are downloaded, converted, and normalized before being stored in a specified path accessible by the WRF model.

[0127] Step 2: Automatically optimize the configuration of 10 schemes in the parameterized scheme library or expand other schemes.

[0128] In the various physical parameterization processes of the WRF model, boundary layer, microphysics, land surface, and near-surface scenarios have a significant impact on wind speed prediction. To systematically evaluate the performance of different scenario combinations, this study constructed and adopted a parameterization scheme library containing 10 specific combinations (Table 1). This library uses multiple schemes such as Kessler, WSM6, WDM5, PurdueLin, Thompson, and Eta in microphysics processes; combinations such as RRTM-Dudhia, RRTMG-RRTMG, and RRTMG-Goddara in radiation processes; Noah, five-layer thermal diffusion, and RUC schemes in land surface processes; MM5, Eta, and MYNN in near-surface scenarios; and YSU, MYNN, QNSE, BouLac, ACM2, and MYJ in boundary layer scenarios. All combinations employ the Kain-Fritsch cumulus convection scheme in the outermost and second nested layers, and disable it in the innermost nested layer to support explicit convection simulation. This example verification mainly revolves around this fixed scheme library to clearly present the scheme selection process. Table 1 shows the results of 10 parameterized combination schemes (scheme1~scheme10) used in the verification process.

[0129] Table 1

[0130] Combination scheme name Microphysical process scheme Longwave / shortwave radiation scheme Land Surface Processes Scheme Near-ground scheme Planetary boundary layer scheme 1 Kessler RRTM-Dudhia Noah MM5 YSU 2 WSM6 RRTMG-RRTMG Noah MYNN MYNN 3 WDM5 RRTMG-RRTMG 5-layer MM5 QNSE 4 PurdueLin RRTM-Dudhia Noah MM5 BouLac 5 Thompson RRTMG-RRTMG Noah MM5 ACM2 6 Eta RRTMG-RRTMG 5-layer Eta MYJ 7 WSM6 RRTMG-RRTMG 5-layer MM5 YSU 8 Thompson RRTMG-Goddara RUC Eta MYJ 9 WSM6 RRTMG-RRTMG 5-layer MM5 BouLac 10 WSM6 RRTMG-Goddara Noah Eta MYJ

[0131] Step 3: Dynamically generate nested meshes based on Lambert projection, and use a triple nested domain approach to ensure fine prediction of the core area. Dynamically reset the height of the eta coordinate layer to ensure that the model layer accurately matches the hub height of the wind turbine generator.

[0132] The nesting ratio of the triple nested domains (D01→D02→D03) is 1:3:3. Based on the Lambert projection and centered on the target wind farm coordinates, nested grids are dynamically generated using auxiliary tools to ensure refined prediction of the core area (D03). For the wind farm in hilly areas, the first layer grid is 85×82 with a spacing of 9km, the second layer is 118×118 with a spacing of 3km, and the third layer is 106×109 with a spacing of 1km. Considering the complexity of the terrain around and near the reservoir, 42 layers are set vertically, with 14 layers densified for heights below 300m near the ground. For the wind farm in plains areas, the first layer grid is 86×83 with a spacing of 9km, the second layer is 100×94 with a spacing of 3km, and the third layer is 82×82 with a spacing of 1km. 37 layers are set vertically, with 12 layers densified for heights below 300m near the ground. The process of refining the ETA coordinate layer configuration is as follows: First, a preliminary run is performed to locate the position index of the reference point in the WRF model grid; then, key meteorological and geographical elements such as air pressure and terrain height at the reference point are read, and the actual height corresponding to the initial ETA layer configuration is calculated; based on this, with the wind turbine hub height as the target, an interpolation algorithm is used to inversely calculate and accurately set the ETA layer height parameters of the WRF mode, thereby ensuring that the vertical layer structure in the model is accurately matched with the hub height.

[0133] Step 4: Using a multi-instance WRF parallel computing framework, the above 10 parameterization schemes are run simultaneously. The system automatically schedules computing resources to efficiently complete the numerical simulation experiments of each scheme and extracts the simulation output data of key meteorological elements such as wind speed, wind direction, temperature, and air pressure at different altitude levels (including hub heights) for each scheme.

[0134] Step 5: Match the simulated output data of each scheme with the corresponding anemometer tower observation data. Based on multiple statistical indicators such as mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and wind direction frequency distribution, a multi-dimensional comprehensive evaluation system is constructed. This system normalizes the scores of wind speed, wind direction, temperature, and air pressure, and assigns weights to each element according to its importance in wind energy projects (wind speed has the highest weight, followed by wind direction, and then temperature and air pressure). The final comprehensive score of each parameterized scheme is calculated through weighted linear superposition.

[0135] The simulated outputs of each parameterized scheme are matched with the corresponding anemometer tower observation data with high precision in time and space. This allows for the construction of a comprehensive evaluation system based on multi-dimensional statistical indicators such as mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and wind direction frequency distribution. According to the scoring method defined in this invention, the system performs layer-by-layer, item-by-item quantitative evaluation of four key meteorological elements: wind speed, wind direction, temperature, and air pressure. In the wind speed element scoring, the MAE, RMSE, and CC sub-scores for each altitude level are first integrated. Then, the average score is taken for all altitude levels containing observation data to obtain the comprehensive wind speed element score. In the wind direction element scoring, the wind direction angle error score is calculated based on the absolute angle error and frequency deviation of the 16 wind direction sectors. Wind direction frequency error score And through weighted fusion (angle error weight) Higher than frequency error weight The comprehensive score of wind direction element is obtained. In the temperature and pressure element scoring, the MAE (Major Air Estimate) of temperature and pressure at each altitude level is calculated separately. After normalization and averaging, the comprehensive score of temperature element is obtained. Comprehensive score with air pressure element Finally, appropriate weights (wind speed weight) are assigned based on the actual importance of each meteorological element in the wind energy project. Highest, wind direction weight Secondly, temperature weighting Weight with air pressure (Lower score), the final comprehensive score of each scheme is calculated using a weighted linear superposition method. The specific formula is as follows:

[0136] Calculate the wind speed sub-score for a single height level:

[0137] ;

[0138] Then, the average value is taken for all height layers containing wind speed observation data to obtain the comprehensive score of wind speed element:

[0139] ;

[0140] in, The total number of height layers containing wind speed observation data, where h is the index of a single height layer. , and Let represent the normalized scores of the mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) of the wind speed at the h-th altitude layer for the i-th physical parameterization scheme, respectively.

[0141] The wind direction records during the observation period are statistically analyzed according to 16 sectors. The average absolute angular error of the i-th scheme is:

[0142] ;

[0143] Where K=16 represents the total number of wind direction sectors, and k is the sector index. Let be the average wind direction of the i-th scheme in the k-th sector. The average wind direction observed in the corresponding sector;

[0144] The wind direction angle error score is calculated as follows:

[0145] ;

[0146] In the formula, and These represent the maximum and minimum absolute errors of the wind direction angle for all schemes;

[0147] The wind direction frequency error score is calculated by taking the arithmetic mean of the deviations in the frequency of wind direction occurrence in each sector.

[0148] The weighted average method was used to calculate the comprehensive score of the wind direction element:

[0149] ;

[0150] In the formula, and These are the angle error weight and the frequency error weight, respectively, with the angle error weight being greater than the frequency error weight.

[0151] Calculate the normalized score of single-level atmospheric pressure error Normalized score of temperature error at single height level :

[0152] ;

[0153] ;

[0154] In the formula, and Let represent the average absolute error of air pressure and the average absolute error of temperature at the h-th altitude layer for the i-th parameterization scheme, respectively. and These represent the maximum and minimum MAE values ​​of air pressure at the h-th altitude layer for all parameterization schemes; and These represent the maximum and minimum MAE values ​​for temperature at the h-th altitude layer for all parameterization schemes;

[0155] Then, the average value is taken for all height layers containing observation data to obtain the comprehensive score of air pressure elements. Comprehensive score based on temperature factor :

[0156] ;

[0157] ;

[0158] In the formula, The total number of altitude layers containing both air pressure and temperature observation data, where h is the index of a single altitude layer;

[0159] Weights were assigned to each meteorological element based on its engineering importance, and the final score was calculated using a weighted linear overlay method.

[0160] ;

[0161] In the formula, , , , These represent the comprehensive scores of the i-th parameterization scheme for wind speed, wind direction, temperature, and air pressure, respectively. , , and These represent the weights of wind speed, wind direction, temperature, and air pressure, respectively. Among them, wind speed has the highest weight, followed by wind direction, while temperature and air pressure have the lowest weights.

[0162] To further verify the simulation capabilities of different parameterization schemes for the diurnal variation characteristics of meteorological elements, this study focuses on the near-surface temperature in plain areas. Figure 3a ), air pressure ( Figure 3b ), wind speed 70m ( Figure 3c ) and 120m ( Figure 3d The diurnal variation curves of each scheme were plotted. Analysis shows that the better-performing schemes (such as Scheme9 and Scheme4) can better reproduce the diurnal variations of temperature, air pressure, and wind speed, especially without significant overestimation during the nighttime stable layer phase, demonstrating their advantages in simulating the dynamic and thermodynamic processes of the atmospheric boundary layer. However, some lower-scoring schemes (such as Scheme10, Scheme1, and Scheme2) exhibit problems such as excessively large wind speed amplitudes and overly strong temperature responses during the nighttime to dawn periods, reflecting their shortcomings in turbulent exchange and surface process parameterization. This diurnal variation analysis provides important process verification evidence for the comprehensive evaluation system, enhancing the physical reliability of the scoring results.

[0163] In this example, the scores and comprehensive scores of the 10 parameterization schemes for wind fields in plains and hilly areas are presented in Tables 2a and 2b, respectively, and are arranged according to... The scores have been ranked. Table 2a shows the predicted wind speed results for the hilly wind field, and Table 2b shows the predicted wind speed results for the plain wind field.

[0164] Table 2a

[0165] Combination scheme name Barometric score Temperature rating Wind speed rating Wind direction and frequency rating Overall score 1 0.81 0.00 0.25 0.56 0.34 2 0.40 0.91 0.20 0.50 0.35 3 1.00 0.46 0.68 0.71 0.70 4 0.66 0.78 0.84 0.47 0.74 5 0.42 0.82 0.44 0.32 0.45 6 0.86 0.69 0.55 0.53 0.59 7 0.39 0.88 0.59 0.23 0.53 8 0.60 0.38 0.39 0.63 0.46 9 0.34 1.00 0.94 0.31 0.76 10 0.02 0.67 0.12 0.61 0.26

[0166] Table 2b

[0167] Combination scheme name Barometric score Temperature rating Wind speed rating Wind direction and frequency rating Overall score 1 0.67 0.85 0.75 0.54 0.71 2 0.79 0.40 0.20 0.46 0.33 3 0.70 0.40 0.43 0.00 0.37 4 0.68 1.00 0.98 0.53 0.86 5 0.86 0.60 0.21 0.97 0.47 6 0.06 0.04 0.38 0.61 0.36 7 0.54 0.36 0.59 0.51 0.55 8 0.85 0.52 0.84 0.83 0.81 9 0.35 0.39 0.52 0.50 0.49 10 0.94 0.28 0.19 0.84 0.41

[0168] Based on this scoring ranking, this invention further implements a dynamic optimization mechanism based on the scenario adaptability preset principle. First, the system associates and stores all quantitative scoring results, the specific scenario characteristics of the target wind farm (including terrain type, meteorological conditions, hub height, etc.), and their corresponding parameterized scheme combinations to form a historical dataset. Second, for the same type of scenario, the system selects the highest-scoring schemes and extracts the common features of their key physical process schemes, thereby identifying the optimal scheme combination type for that scenario. For example, in plain areas, the highest-scoring Scheme 9, Scheme 4, and Scheme 3 all exhibit a commonality in their boundary layer schemes, primarily using BouLac or YSU. Simultaneously, the system analyzes the score changes of the same parameterized scheme in different scenarios, summarizing its scenario adaptability preferences. For instance, Scheme 4 scores significantly higher in hilly areas (0.86) than in plain areas (0.74), indicating better adaptability to complex underlying surfaces. Subsequently, the contribution percentage of each element score (wind speed, wind direction, temperature, and air pressure) to the final score was calculated for each scenario, identifying key meteorological elements affecting the suitability of the proposed solutions. For example, in hilly areas, wind speed and air pressure scores contribute significantly, while in plains areas, wind speed and temperature scores have a more prominent impact. Based on the established mapping relationship between scores and solution suitability, the system dynamically adjusts the selection priority of parameterized solutions: physical process solutions corresponding to the common characteristics of high-scoring solutions are prioritized for that type of scenario, scenario-specific suitability rules are supplemented, and the weight of corresponding physical process solutions in the selection is adjusted according to the contribution percentage of each meteorological element. Finally, the optimized scenario suitability preset principle is applied to a new simulation task. The effectiveness of the optimization is verified by comparing the improvement in the overall score of the optimal solution before and after the update, as well as the change in the score differentiation between different solutions. If the optimal solution score is significantly improved and the differentiation between solutions is enhanced under the new principle, the update is confirmed to be effective; otherwise, correlation analysis and iterative optimization are performed again.

[0169] In step S6, based on the comprehensive score ranking results obtained in step 5, the system automatically selects the parameterization scheme with the best overall performance. In this example, the optimal scheme for the plains region (Bozhou, Anhui) is Scheme 9, specifically configured as follows: microphysical process WSM6, longwave radiation RRTMG and shortwave radiation RRTMG, land surface process 5-layer, near-surface scheme MM5, and boundary layer scheme BouLac. This scheme achieves a comprehensive score of 0.76 in the plains region, demonstrating the best performance in reproducing diurnal wind speed variations and coordinating multiple elements. The optimal scheme for the hilly region (Dalian, Liaoning) is Scheme 4, specifically configured as follows: microphysical process PurdueLin, longwave radiation RRTM and shortwave radiation Dudhia, land surface process Noah, near-surface scheme MM5, and boundary layer scheme BouLac. This scheme achieves a comprehensive score of 0.86 in the hilly region, particularly excelling in the accuracy of wind speed and air pressure prediction. The average score for both regions reaches 0.80, ranking first among all schemes (see Table 3), demonstrating excellent cross-scene adaptability.

[0170] After selecting the optimal solution, the system restarts the WRF mode with all the physical processes configured for that solution and executes the final forecast simulation. This simulation continues to use the triple-nested grid system optimized in step 3 (with the innermost layer having a resolution of 1 km) and a vertical layer structure with increased density near the hub height to ensure that the model has high-resolution characterization capabilities in key areas. After the simulation is completed, the system outputs two types of core results that can be directly used for engineering decision-making: the first is the hourly wind speed and direction sequence that perfectly matches the wind turbine hub height. This result can be directly input into the wind power curve model for real-time calculation and short-term prediction of wind farm power generation. The second is high-resolution three-dimensional wind field and meteorological field data covering multiple height levels in the vertical direction, including elements such as wind speed, wind direction, temperature, and air pressure from near the ground to above the hub height. Its horizontal resolution is consistent with the innermost simulation grid, thus supporting various engineering applications such as wind turbine layout optimization, wake impact assessment, extreme wind condition identification, and fine analysis of wind resource spatial distribution. Through the aforementioned fully automated, closed-loop optimized forecasting process, this invention significantly improves the accuracy, automation level, and engineering applicability of wind resource forecasting using the WRF model under different geographical and meteorological scenarios, providing a reliable data foundation for wind farm planning, operation, and maintenance. Table 3 shows a comprehensive comparison of various parameter schemes for the two wind farms.

[0171] Table 3

[0172] Combination scheme name Comprehensive score of the calculation example in Qiaocheng, Anhui Comprehensive score of the case study of Liuda Reservoir in Liaoning Average rating Preferred ranking 1 0.34 0.71 0.525 6 2 0.35 0.33 0.340 9 3 0.70 0.37 0.535 5 4 0.74 0.86 0.800 1 5 0.45 0.47 0.460 8 6 0.59 0.36 0.475 7 7 0.53 0.55 0.540 4 8 0.46 0.81 0.635 2 9 0.76 0.49 0.625 3 10 0.26 0.41 0.335 10

[0173] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0174] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization, characterized in that, The method includes: S1 automatically acquires GFS meteorological driving data, geographic static data and high-precision terrain data of the target wind farm, performs integrity verification and format standardization on the acquired data, and obtains input field data for subsequent simulation. S2, construct a fixed parameterization scheme library containing a set of typical physical parameterization combination schemes, and select whether to randomly generate parameterization extension schemes based on the underlying surface and meteorological conditions data of the target wind farm or the control commands input by the user. If so, according to the requirements of wind farm engineering application for the simulation accuracy and stability of key meteorological elements, and in accordance with the scenario adaptability preset principle, automatically parameterize the key physical processes of WRF mode to generate multiple extended parameterization schemes, and proceed to step S3; otherwise, directly proceed to step S3. S3. Based on the location of the forecast points, dynamically determine the WRF nesting scheme, which includes the number of nesting layers, the grid resolution and range of each layer; execute the WPS module in sequence to generate the initial field and boundary conditions of the model; dynamically adjust the vertical layer height of the eta coordinate to make the model layer match the hub height of the wind turbine, and refine the vertical layer near the corresponding hub height. S4 employs a multi-instance WRF parallel computing framework to simultaneously run fixed parameterization schemes and extended parameterization schemes, acquiring simulation data of key meteorological elements, including wind speed, temperature, air pressure, and wind direction, under each scheme. S5 performs spatiotemporal matching and processing of key meteorological element simulation data and corresponding observation data. Based on mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and multiple meteorological indicators including wind direction frequency distribution, a comprehensive evaluation system is constructed to quantitatively score the simulation performance of each parameterized scheme and adjust the scenario adaptability preset principle according to the quantitative score. S6 quantifies and ranks the simulation performance of each parameterization scheme, automatically selects the parameterization scheme with the best overall performance based on the ranking results, and restarts the WRF mode with the parameterization scheme to perform the final forecast simulation, outputting hub height and multi-level high-resolution wind field data. Step S5 further includes: S51, Calculate the wind speed sub-score for a single height level. : ； Then, the average value is taken for all height layers containing wind speed observation data to obtain the comprehensive score of wind speed element. : ； in, The total number of height layers containing wind speed observation data, where h is the index of a single height layer. , and Let represent the normalized scores of the mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC) of the wind speed at the h-th altitude layer for the i-th physical parameterization scheme, respectively. S52, the wind direction records during the observation period are statistically analyzed according to 16 sectors, and the average absolute angular error of the i-th scheme is... for: ； Where K=16 represents the total number of wind direction sectors, and k is the sector index. Let be the average wind direction of the i-th scheme in the k-th sector. The average wind direction observed in the corresponding sector; Calculate the wind direction angle error score : ； In the formula, and These represent the maximum and minimum absolute errors of the wind direction angle for all schemes; S53, after taking the arithmetic mean of the deviations in the frequency of wind direction occurrence in each sector, the wind direction frequency error score is calculated. ; S54 uses a weighted average method to calculate the comprehensive score of wind direction element. : ； In the formula, and These are the angle error weight and the frequency error weight, respectively, with the angle error weight being greater than the frequency error weight. S55, Calculate the normalized score of single-level atmospheric pressure error. Normalized score of temperature error at single height level : ； ； In the formula, and Let represent the average absolute error of air pressure and the average absolute error of temperature at the h-th altitude layer for the i-th parameterization scheme, respectively. and These represent the maximum and minimum MAE values ​​of air pressure at the h-th altitude layer for all parameterization schemes; and These represent the maximum and minimum MAE values ​​for temperature at the h-th altitude layer for all parameterization schemes; Then, the average value is taken for all height layers containing observation data to obtain the comprehensive score of air pressure elements. Comprehensive score based on temperature factor : ； ； In the formula, The total number of altitude layers containing both air pressure and temperature observation data, where h is the index of a single altitude layer; S56, weights are assigned according to the engineering importance of each meteorological element, and the final score is calculated using a weighted linear overlay: ； In the formula, , , , These represent the comprehensive scores of the i-th parameterization scheme for wind speed, wind direction, temperature, and air pressure, respectively. , , and These represent the weights of wind speed, wind direction, temperature, and air pressure, respectively. Among them, wind speed has the highest weight, followed by wind direction, while temperature and air pressure have the lowest weights.

2. The automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization according to claim 1, characterized in that, Step S1 further includes: Determine the latitude and longitude of the prediction point and the target prediction time period for the numerical simulation of the target wind farm; based on the set target prediction time period, set the start and end times of the meteorological driving field, and support automatic downloading at four times: UTC00, 06, 12, and 18. Download SRTM3s resolution elevation data tiles near the prediction point, merge the hgt format data and convert it to tif format, and then further convert it to geogrid binary format; then create a folder named srtm_3s under the geog_data_path path in the namelist.wps configuration file, and place the processed elevation data and index files in this folder directory for subsequent modeling and simulation.

3. The automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization according to claim 1, characterized in that, In step S2, based on the requirements of wind farm engineering applications for the simulation accuracy and stability of key meteorological elements, and following the principle of scenario adaptability, the process of automatically parameterizing the key physical processes of the WRF mode and generating multiple extended parameterization schemes includes: The simulation accuracy and stability requirements of key meteorological elements for wind farm engineering applications, including hub height wind speed error and wind direction prediction deviation, are quantitatively analyzed and matched with the key physical processes of the WRF model. The key physical processes of the WRF model include microphysical processes, long-wave and short-wave radiation processes, boundary layer turbulent mixing mechanism, near-surface layer flux exchange process, cumulus convection process and land surface process. Based on the matching results, and in accordance with the principle of scenario adaptability, one or more candidate schemes for each key physical process are selected from the preset scheme parameter library. Based on the parameter combination of the fixed parameterization scheme, and with the core constraint of maintaining the forced coordination between the boundary layer and near-ground layer schemes, multiple sets of extended parameterization schemes with differentiated configurations are generated by replacing single physical process schemes or replacing single sets of coordinated physical process schemes. Specifically, for key physical processes that are not coordinated, the single physical process scheme replacement method is adopted, and for physical process schemes with coordinated logic, the single set of coordinated physical process scheme replacement method is adopted.

4. The automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization according to claim 1, characterized in that, Step S3 further includes: Based on the latitude and longitude coordinates of the target wind farm, a Lambert conformal conic projection method is adopted, and a triple-nested grid system is constructed according to a horizontal nesting ratio of 1:3:

3. The outermost grid covers the mesoscale region where the target wind farm is located, and is used to introduce the large-scale circulation background. The middle layer covers the regional scale where the target wind farm is located, and is used to achieve a smooth transition from large-scale to small-scale meteorological fields. The innermost high-resolution core domain directly covers the target wind farm area to support the fine simulation of the wind farm. In the vertical direction, the terrain-following eta coordinate system is adopted, and the total number of vertical layers in the whole area is set. The vertical layer densification is carried out near the height of the wind turbine hub through an adaptive optimization algorithm. The layer spacing of the layers above and below the hub height is smaller than that of other layers, so that the hub height area has higher vertical resolution to characterize the wind speed profile, turbulence structure and boundary layer thermodynamic characteristics at the hub height.

5. The automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization according to claim 1, characterized in that, Step S5, the process of adjusting the preset principles for scene adaptability based on the quantitative score, includes: The quantitative scoring results, target wind farm scene features, and corresponding parameterization schemes are combined and stored together to form a structured dataset; scene features, including terrain type, meteorological conditions, and hub height, are quantified according to preset categories. For the same type of scenario, the top-scoring parameterized schemes are selected, and the common characteristics of their key physical process schemes are extracted to identify the optimal scheme combination type for that scenario. Then, the score changes of the same parameterized scheme in different scenarios are analyzed to obtain the scenario adaptation preference of the parameterized scheme. Finally, for each scenario, the contribution ratio of the comprehensive score of wind speed, wind direction, temperature, and air pressure to the final score is calculated to determine the key meteorological factors affecting the suitability of the scheme in each scenario, and the mapping relationship between the score and the suitability of the parameterized scheme is analyzed. Based on the mapping relationship between scores and the adaptability of parameterized schemes, the priority of parameterized scheme selection is adjusted, and the physical process schemes corresponding to the common characteristics of high-scoring schemes are listed as the priority selection options for the corresponding scenarios; at the same time, scenario-specific adaptation rules are added to clarify the optimal scheme combination type of key physical processes under different scenarios; and the selection weight of corresponding physical process schemes is adjusted according to the contribution ratio of meteorological elements. Apply the adjusted scenario adaptability preset principle to the new simulation task and compare the changes in the overall score before and after the update. If the overall score of the optimal solution improves under the new principle and the score differentiation of different solutions improves, then the update is confirmed to be effective; otherwise, re-analyze the correlation between the quantitative score and the parameterized solution.

6. The automatic optimization wind resource forecasting method based on multi-parameter scheme selection and WRF nested optimization according to claim 1, characterized in that, In step S5, the timing of adjusting the scene adaptability preset principle is controlled by setting update trigger conditions.

Images