Methods and systems for correcting wind speed deviations by fusing anemometer tower data with reanalysis data

Abstract

This invention provides a method, system, electronic device, and storage medium for wind speed deviation correction by fusing anemometer data and reanalysis data. First, it acquires multi-height anemometer data, reanalysis data, topographic and surface feature data, and satellite data for the target wind resource assessment area. Using the near-ground reference height wind speed from the reanalysis data as an anchor point, it combines topographic roughness calculated from the DEM with dynamic roughness extrapolated from seasonal NDVI / LAI. Based on PBLH, cloud cover, and diurnal information, stability is categorized, and both are integrated into a time-varying shear index construction. A physically consistent baseline wind speed is obtained through power-law extrapolation. Subsequently, the standardized dataset and baseline wind speed are integrated into an hourly feature vector containing multi-dimensional information. The difference between the anemometer measured wind speed and the baseline wind speed is used as the residual label to train a supervised prediction model capable of predicting residuals and uncertainties. Finally, the residuals for the target area are predicted, and the baseline is superimposed to obtain the preliminary corrected wind speed. After multi-dimensional calibration, the final corrected wind speed and uncertainty confidence interval are output.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of wind energy resource assessment and meteorological data fusion technology, and in particular to a method, system, electronic device and storage medium for wind speed deviation correction by fusing wind tower data and reanalysis data. Background Technology

[0002] Reanalysis data (such as ERA5 (ECMWF Re-Analysis 5, ECMWF Re-Analysis 5-Land, and MERRA-2) are widely used in wind energy resource assessment due to their high spatial resolution and complete temporal coverage. However, in wind speed simulations at heights of 100m and 150m, these reanalysis data exhibit significant systematic biases due to factors such as complex terrain, abrupt changes in surface roughness, and variations in boundary layer stability, directly leading to inaccurate wind energy resource assessment results.

[0003] Existing wind speed deviation correction techniques mostly rely on traditional linear methods such as ratio methods and power-law extrapolation, which cannot effectively handle the complex nonlinear relationships between wind speed and topography and stability. This is especially true in scenarios where wind speed and topography interact strongly, leading to significantly increased errors. Furthermore, existing methods often focus on energy / power prediction or spatial interpolation, generally neglecting the physical consistency of data from multiple height meteorological towers during the fusion of reanalysis data. They lack specific wind speed deviation correction schemes for the 100-meter height level; moreover, there are no effective means to quantify the uncertainty of the correction results, making it difficult to provide operable confidence intervals for wind speed assessment. They also suffer from insufficient transferability in small sample sites. All these shortcomings limit the accuracy and reliability of wind energy resource assessment. Summary of the Invention

[0004] This invention provides a wind speed deviation correction method, system, electronic device, and storage medium for fusing wind tower data and reanalysis data, in order to solve the technical problems in the prior art, such as the systematic deviation of wind speed at the 100-meter level in reanalysis data, the inability of linear correction methods to effectively handle the nonlinear relationship between wind speed and topography and stability, the insufficient migration and adaptation capability of small sample sites, and the lack of effective means to quantify the uncertainty of wind speed correction results.

[0005] In a first aspect, embodiments of the present invention provide a method for correcting wind speed deviations by fusing anemometer data and reanalysis data, comprising: S1. Obtain multi-height wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and perform spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. S2. Using the near-Earth reference height wind speed from the reanalysis data as an anchor point, dynamic roughness is obtained. Based on the stability classification according to planetary boundary layer height (PBLH), cloud cover, and day / night information, dynamic roughness and stability classification are integrated into the construction process of the time-varying shear index to obtain the time-varying shear index that changes dynamically over time. Based on the power-law extrapolation formula, using the time-varying shear index as the core parameter, the near-Earth reference height wind speed is extrapolated to the target height to obtain the physically consistent baseline wind speed. S3. Integrate the standardized dataset and the physical consistency baseline wind speed into hourly feature vectors. Use the difference between the wind speed measured by the wind tower at the same time and the baseline wind speed as the residual label. Construct a sample set in which the hourly feature vectors and the residual labels of the same period correspond one-to-one. Train a supervised prediction model that can predict wind speed residuals and uncertainties based on the sample set. S4. Based on the wind speed residual of the target wind resource assessment area predicted by the supervised prediction model, the wind speed residual is superimposed with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed. Then, physical calibration is performed through high monotonicity constraints, roughness consistency smoothing and extreme wind speed limits to obtain the final corrected wind speed and the corresponding uncertainty confidence interval.

[0006] Preferably, in step S1, the multi-height-layer wind tower data includes wind speed data with a continuous observation period of at least one year, the resolution of the wind speed data is 10 minutes, the wind speed data covers at least two hundred-meter-level heights of 100m and 150m, and the multi-height-layer wind tower data also includes temperature, humidity, and air pressure. The reanalysis data includes wind speed, air temperature, air pressure, humidity, friction speed, stability, and cloud cover; The terrain and surface feature data include digital elevation model (DEM) data, and terrain roughness data, slope data, and aspect data derived from the DEM data, as well as seasonal normalized vegetation index (NDVI) data and leaf area index (LAI) data. Step S1 also includes acquiring satellite data of the target wind resource assessment area, including cloud cover and shortwave radiation parameters.

[0007] Preferably, in step S2, obtaining the dynamic roughness includes: Calculate terrain roughness based on the Digital Elevation Model (DEM):

[0008] In the above formula, R is the terrain roughness index, in meters (m). z i For the first DEM window i Elevation values ​​of each grid point; The average elevation of the DEM window; n This represents the number of grid points within the DEM window. Dynamic roughness is calculated by combining the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) using an empirical function.

[0009] In the above formula, z 0,terrain The basic terrain roughness data obtained from DEM calculation, a , b These are empirical coefficients fitted to historical wind measurement tower data. t It is a time variable; NDVI ( t )for t Normalized Difference Vegetation Index (LAI) at any given time t )for t Leaf area index at any given time.

[0010] Preferably, in step S2, the specific formula for constructing the time-varying shear index is as follows:

[0011] In the above formula, α 0 is the base coefficient; c 1~ c 4 represents the coefficients used to fit the historical meteorological tower data; S ( t The stability is categorized into levels, with -1 for unstable, 0 for neutral, and +1 for stable; PBLH is the planetary boundary layer height; the stability categorization is based on the following criteria: strong solar radiation during the day and a thick PBLH indicate unstable conditions, radiative cooling at night and a shallow PBLH indicate stable conditions, and all other conditions are classified as neutral.

[0012] Preferably, in step S3, the hourly feature vector includes wind speed difference, wind direction sine and cosine coding, terrain upwind index, stability grading, PBLH and cloud cover parameters in the reanalysis data, and data quality markers; wherein, the wind speed difference is the wind speed difference between the target height and the near-ground reference height; the terrain upwind index is the difference between wind direction and slope aspect; and the data quality markers include complement markers and low-trust markers.

[0013] Preferably, in step S4, when the wind speed residual of the target wind resource assessment area is predicted according to the supervised prediction model, if it is determined that the wind speed data of the multi-height-level wind measurement towers in the target wind resource assessment area has less than 3 months of continuous observation time, a source domain subgroup with a similarity of ≥80% to the climate and topographic features of the target domain is selected from the accumulated multiple complete data stations. The kernel mean matching (KMM) method is adopted, and the hourly feature vector constructed in S3 is used as the feature dimension to adjust the sample weights of the source domain subgroup, so that the Wasserstein distance between the adjusted source domain feature distribution and the feature distribution of a small number of samples in the target wind resource assessment area is ≤0.2, thus eliminating the distribution offset. The tree structure parameters of the supervised prediction model trained by S3 are frozen. Based on the multi-height wind measurement tower data of the target wind resource assessment area, the output bias of the leaf nodes of the supervised prediction model is fine-tuned with L2 regularization. The regularization coefficient is... λ =0.01, so that the residual prediction MAE of the supervised prediction model in the target domain with few samples after fine-tuning is reduced by ≥15% compared with that before fine-tuning. After completing the cross-site migration adaptation, the wind speed residual prediction of the target wind resource assessment area is then carried out.

[0014] Preferably, in step S4, physical calibration is further performed through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limiting, including: An equidistant regression algorithm is used, with the initial corrected wind speeds at heights of 100m and 150m as inputs. If the initial corrected wind speed at 150m is less than the initial corrected wind speed at 100m, or if the wind speed at either height is less than 0m / s, the wind speeds at the two heights are adjusted synchronously through equidistant regression. After adjustment, the final corrected wind speed at 150m must be greater than or equal to the final corrected wind speed at 100m, and the wind speeds at both heights must be greater than or equal to 0m / s. The absolute deviation between the adjusted wind speed and the initial corrected wind speed must be less than or equal to 0.5m / s. Based on the terrain roughness in the terrain and surface feature data, the roughness difference between adjacent grid points is calculated. If the roughness difference exceeds the preset range threshold, the corrected wind speed in the roughness transition area is smoothed by Gaussian filtering. Wind speed upper limit thresholds are set according to different climate zones and stability conditions. When the predicted wind speed exceeds the upper limit threshold, it is truncated and marked. The mark needs to be output along with the final corrected wind speed.

[0015] Secondly, an embodiment of the present invention provides a wind speed deviation correction system that fuses wind measurement tower data with reanalysis data, comprising: The data acquisition and preprocessing module acquires multi-height-level wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and performs spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. The baseline wind speed construction module uses the near-Earth reference height wind speed from reanalysis data as an anchor point to obtain dynamic roughness. Based on the stability classification according to planetary boundary layer height (PBLH), cloud cover, and day / night information, the dynamic roughness and stability classification are integrated into the construction process of the time-varying shear index to obtain the time-varying shear index that changes dynamically over time. Based on the power-law extrapolation formula, the near-Earth reference height wind speed is extrapolated to the target height using the time-varying shear index as the core parameter to obtain the physically consistent baseline wind speed. The feature integration and model training module integrates the standardized dataset and the physically consistent baseline wind speed into hourly feature vectors. The difference between the wind speed measured by the wind tower at the same time and the baseline wind speed is used as the residual label. A sample set with one-to-one correspondence between the hourly feature vectors and the residual labels of the same period is constructed. A supervised prediction model that can predict wind speed residuals and uncertainties is trained based on the sample set. The residual prediction and physical calibration module predicts the wind speed residual of the target wind resource assessment area based on the supervised prediction model. It then superimposes the wind speed residual with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed. Finally, it performs physical calibration through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limits to obtain the final corrected wind speed and the corresponding uncertainty confidence interval.

[0016] Thirdly, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the wind speed deviation correction method for fusing wind tower data and reanalysis data as described in the first aspect of the present invention.

[0017] Fourthly, embodiments of the present invention provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the wind speed deviation correction method for fusing wind tower data and reanalysis data as described in the first aspect of the present invention.

[0018] This invention provides a method, system, electronic device, and storage medium for correcting wind speed deviations by fusing anemometer data and reanalysis data. The core logic is the fusion of physical constraints and machine learning. First, it acquires multi-height anemometer data, reanalysis data, topographic and surface feature data, and satellite data for the target wind resource assessment area. A standardized dataset is obtained through spatiotemporal matching, three-layer outlier removal, hierarchical missing value filling, and three types of consistency checks. Then, using the near-ground reference height wind speed from the reanalysis data as an anchor point, it combines topographic roughness calculated from the DEM with seasonal NDVI / LAI to infer dynamic roughness, based on PBLH, cloud cover, and... By classifying day and night data into stability tiers and integrating them into a time-varying shear index, a physically consistent baseline wind speed is obtained through power-law extrapolation. Subsequently, a standardized dataset and the baseline wind speed are integrated into an hourly feature vector containing multi-dimensional information. The difference between the measured wind speed from the wind tower and the baseline wind speed is used as the residual label to train a supervised prediction model capable of predicting residuals and uncertainties. Finally, the residuals are predicted for the target area (for stations with few samples, climate zone gating and KMM reweighting are required for migration adaptation), and the baseline is superimposed to obtain the preliminary corrected wind speed. After high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limiting calibration, the final corrected wind speed and uncertainty confidence interval are output. This scheme effectively solves the problems of systematic bias in wind speed at the 100-meter level, difficulty in handling nonlinear relationships with linear methods, insufficient migration capability for few samples, and lack of uncertainty quantification methods in existing technologies. It significantly improves wind speed correction accuracy, adapts to different climate zones and complex terrains, and meets the accuracy and reliability requirements of wind power planning projects. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a wind speed deviation correction method that integrates wind measurement tower data and reanalysis data according to an embodiment of the present invention. Figure 2 This is a block diagram of a wind speed deviation correction system that integrates wind tower data and reanalysis data according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the physical structure according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0023] The terms "first" and "second" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a system, product, or device that includes a series of components or units is not limited to the listed components or units, but may optionally include unlisted components or units, or may optionally include other components or units inherent to such products or devices. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] This invention provides a method for correcting wind speed deviations by fusing anemometer data and reanalysis data, such as... Figure 1 As shown, it includes: S1. Obtain multi-height wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and perform spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. S2. Using the near-Earth reference height wind speed from the reanalysis data as an anchor point, dynamic roughness is obtained. Based on the stability classification according to planetary boundary layer height (PBLH), cloud cover, and day / night information, dynamic roughness and stability classification are integrated into the construction process of the time-varying shear index to obtain the time-varying shear index that changes dynamically over time. Based on the power-law extrapolation formula, using the time-varying shear index as the core parameter, the near-Earth reference height wind speed is extrapolated to the target height to obtain the physically consistent baseline wind speed. S3. Integrate the standardized dataset and the physical consistency baseline wind speed into hourly feature vectors. Use the difference between the wind speed measured by the wind tower at the same time and the baseline wind speed as the residual label. Construct a sample set in which the hourly feature vectors and the residual labels of the same period correspond one-to-one. Train a supervised prediction model that can predict wind speed residuals and uncertainties based on the sample set. S4. Based on the wind speed residual of the target wind resource assessment area predicted by the supervised prediction model, the wind speed residual is superimposed with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed. Then, physical calibration is performed through high monotonicity constraints, roughness consistency smoothing and extreme wind speed limits to obtain the final corrected wind speed and the corresponding uncertainty confidence interval.

[0026] Specifically, in step S1, the multi-height-layer wind tower data includes wind speed data with a continuous observation period of at least one year. The resolution of the wind speed data is 10 minutes, and the wind speed data covers at least two hundred-meter-level heights of 100m and 150m. The multi-height-layer wind tower data also includes temperature, humidity, and air pressure. After outlier removal, missing value filling, and consistency verification, the data quality is ensured.

[0027] The reanalysis data includes wind speed, air temperature, air pressure, humidity, friction velocity, stability, and cloud cover; in this embodiment, spatiotemporally matched reanalysis data such as ERA5-Land are selected.

[0028] The terrain and surface feature data include digital elevation model (DEM) data, and terrain roughness data, slope data, and aspect data derived from the DEM data, as well as seasonal normalized vegetation index (NDVI) data and leaf area index (LAI) data.

[0029] Step S1 also includes acquiring satellite data of the target wind resource assessment area. In this embodiment, satellite data such as FY-4B are introduced, and the satellite data includes cloud cover and shortwave radiation parameters.

[0030] The data also needs to be preprocessed, specifically including: 1) Spatiotemporal matching: Unified timeline: All records are aligned to the nearest 10-minute time; Deduplication (if multiple records are taken at the same time, the first record is taken and the rest are marked as "duplicate"); Missing time slots are filled; Time zone and time scale: Unified to local time or UTC (consistent with reanalysis); If the station clock drift is >±60s, the entire system is shifted and aligned; Units and precision: Wind speed m / s, temperature ℃, relative humidity %, air pressure hPa; Decimal places are unified to 1 place for wind speed, 1 place for temperature and humidity, and 1 place for air pressure.

[0031] 2) Outlier Removal: Outlier removal is performed using a three-tiered system: "hard threshold → statistical threshold → behavioral characteristics". Exceeding limits: Wind speed: <0 or >60m / s; Temperature: <-50 or >50℃; Relative humidity: <0 or >100%; Air pressure: <500 or >1100hPa.

[0032] Gross error: deviation from the neighborhood median within a 10-minute window > 3 times the "absolute neighborhood deviation".

[0033] Abrupt changes: wind speed difference > 20 m / s; temperature difference > 5℃; humidity difference > 20%; air pressure difference > 3 hPa.

[0034] Constant value data: ≥12 consecutive points (≥2h) are completely identical, or variance = 0; Wind direction consistency: first convert the wind direction to u / v component to detect peaks and interpolation; false abrupt changes caused by 0 / 360° crossings are not counted as anomalies.

[0035] 3) Missing value imputation It is only used for gaps to maintain sequence continuity; any padded values ​​are marked with "Insufficiency Marker (IMP)" and can be used as features during training.

[0036] Level A (linear interpolation, priority): For single-element single-height continuous gaps ≤30min (≤3 points), linear interpolation is performed when both ends are valid; wind direction is first interpolated on the u / v components and then the angle is restored.

[0037] Level B (Diurnal Median Backfill): 30–180 min gap, backfilled according to the historical median of "same month, same hour (±1 hour), and the last ±15 days"; wind direction is represented by the vector median.

[0038] Level C (Reference Field Driven Regression): Gap >180 min and ≤6 h, backfilled using historical linear relationships with reanalysis (e.g., ERA5 / ERA5-Land) or nearby qualified heights; residuals are limited to ±1.5 times the historical IQR for that period, otherwise left blank.

[0039] Long gaps exceeding 6 hours: Do not fill, retain the missing information, and remove this period during the training and evaluation phases.

[0040] Aggregated to the hour, at least 4 valid 10-minute samples per hour are required to provide an hourly value; wind speed is averaged, and wind direction is vector averaged; hours with a complement ratio >50% are marked as "low trust".

[0041] 4) Consistency check Temporal consistency: All four elements must be present or marked as missing at the same time; if only wind speed is present while meteorological elements are missing for a long period of time, it will only be used for limited tasks and marked as "element missing".

[0042] Vertical consistency (physical plausibility, not hard exclusion): Wind speed at 150m should generally not be less than 100m. If 150m < 100m... Points with a wind speed of 1.0 m / s and sustained for ≥3 consecutive points are marked as "headwind shear layer" and are not directly deleted, but left to be interpreted by the model using stability characteristics.

[0043] Physical constraints of elements: oRH is 0–100% (rounding error is allowed to be ±1%). The dew point is not higher than the temperature (if the dew point can be calculated); Air pressure and altitude are consistent for a short period of time (a long-term systematic deviation indicates that the pressure sensor needs maintenance).

[0044] Furthermore, in S2, the near-ground reference height wind speed adopts the 5m-15m height wind speed from the reanalysis data, preferably the 10m height wind speed. This height is the standard output parameter of the reanalysis data (such as ERA5-Land) and is minimally affected by complex terrain at the hundred-meter level and atmospheric boundary layer stability. The data deviation is controllable, so it is used as an anchor point (i.e., benchmark reference data) to ensure the reliability of the starting point for subsequent extrapolation.

[0045] In step S2, the wind speed at a near-ground reference height of 5m-15m in the reanalysis data is selected as the anchor point to ensure the reliability of the starting data. Based on the topographic roughness derived from the DEM, dynamic roughness is derived through an empirical function using seasonal NDVI / LAI to quantify the dynamic frictional influence of topography and vegetation. Stability is categorized into (-1 / 0 / +1) based on PBLH, cloud cover, and diurnal information to quantify the dynamic influence of the atmosphere on the vertical distribution of wind. Dynamic roughness, stability grading, and PBLH constraints are substituted into the fitting formula to obtain the time-varying shear index, establishing the correlation between the surface-atmosphere and wind speed distribution. Power-law extrapolation: using the time-varying shear index... α(t) Using the near-ground reference wind speed as the core parameter, the wind speed is extrapolated to a target height of 80m-200m to obtain the physically consistent baseline wind speed. This breaks through the traditional method of fixing the shear index (e.g., αOvercoming the limitations of (=0.14), this method adapts the baseline wind speed to the dynamic changes of terrain, vegetation, and stability by using a time-varying shear index. This ensures the baseline wind speed conforms to the physical laws of the atmospheric boundary layer, avoiding simulation deviations at the hundred-meter height level caused by fixed parameters. The baseline wind speed incorporates both anchoring information from the reanalysis data (ensuring data continuity) and physical dynamic factors (ensuring authenticity). The residuals between the subsequent anemometer tower measurements and the baseline accurately reflect the systematic biases of the reanalysis data, laying a high-quality foundation for machine learning residual learning. Dynamic roughness covers terrain undulations and seasonal vegetation changes, while stability is graded to adapt to atmospheric conditions at different times. This allows the baseline wind speed to be adapted to complex terrains such as mountains and coastlines, as well as different climate zones (such as monsoon regions and temperate zones), improving the method's versatility.

[0046] Furthermore, in S3, by integrating high-quality data from the preprocessing stage with the physical baseline, the multi-dimensional information affecting wind speed deviation is transformed into features that the model can learn. Using the deviation between the measured data and the baseline as the learning target, a model that can accurately predict wind speed residuals and uncertainties is trained, providing a data-driven deviation correction tool for subsequent wind speed correction in the target area, thus solving the problem that traditional methods cannot quantify nonlinear deviations.

[0047] Furthermore, in S4, reanalysis data, topographic and surface feature data (including DEM / NDVI / LAI) of the target area are acquired. After preprocessing in S1, hourly feature vectors of the same origin as those in S3 are constructed. If it is a new site, the model is optimized by gating climate zones → reweighting KMM samples → light fine-tuning QRF leaf nodes to adapt to the characteristics of the target domain. The feature vectors are input into the QRF model to obtain the predicted residual point values ​​(p50) and p05 / p95 quantiles. The predicted residual point values ​​are superimposed with the physical consistency baseline wind speed in S2 to generate the preliminary corrected wind speed. The high monotonicity constraint (isotrace regression), roughness consistency smoothing (Gaussian filtering), and extreme wind speed limiting (thresholds set according to climate zone / stability) are executed in sequence to eliminate physical inconsistencies. Combining the residual p05 / p95 quantiles with the calibrated final wind speed, a confidence interval of PICP ≥ 80% is obtained through conformal calibration. Finally, the final corrected wind speed + uncertainty confidence interval + quality label (such as extreme limiting label) is output.

[0048] By employing a migration scheme that combines climate zone gating, KMM reweighting, and lightweight fine-tuning, the model can still be stably corrected even when the new site contains only two months of data. This overcomes the limitation of existing technologies that rely on long-term measured data and expands the applicability of the method.

[0049] Based on the above embodiments, as a preferred implementation, in step S2, obtaining the dynamic roughness includes: Calculate terrain roughness based on the Digital Elevation Model (DEM):

[0050] In the above formula, R is the terrain roughness index, in meters (m). z i For the first DEM window i Elevation values ​​of each grid point; The average elevation of the DEM window; n This represents the number of grid points within the DEM window (e.g., 3×3 or 5×5 pixels).

[0051] Dynamic roughness is calculated by combining the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) using an empirical function.

[0052] In the above formula, z 0,terrain The basic terrain roughness data obtained from DEM calculation, a , b These are empirical coefficients fitted to historical wind measurement tower data. t It is a time variable; NDVI ( t )for t Normalized Difference Vegetation Index (LAI) at any given time t )for t Leaf area index at any given time.

[0053] Specifically, the Digital Elevation Model (DEM) also extracts the following terrain features: Slope refers to the maximum angle of inclination of the earth's surface at a given point. It is calculated from the elevation values ​​of a DEM using gradient gradation.

[0054] In the above formula, S is the slope, in degrees or rad (radians); z is the elevation value of a point in the DEM, in meters. It represents the rate of change of elevation along the east-west direction (y-direction, longitude direction), that is, the difference in elevation per unit distance in the horizontal direction; is the rate of change of elevation in the north-south direction (y-direction, latitude direction); arctan() is the arctangent function, which converts the rate of change of elevation into an angle. The greater the slope, the steeper the terrain, and the more significant the disturbance to the wind.

[0055] Aspect calculation refers to the azimuth angle of the direction of maximum slope relative to true north. The formula is:

[0056] A represents the slope aspect, usually expressed as 0°-360°, where 0° is due north and increases clockwise; arctan2(y,x) is a two-parameter arctangent function that guarantees the result within the complete 0°→360° range. It is the negative value of the north-south elevation gradient to ensure that the direction is consistent with the actual terrain; This represents the east-west elevation gradient. Slope aspect determines the distribution of the windward and leeward sides of a mountain, and has a crucial impact on the acceleration or deceleration of wind speed.

[0057] Based on the above embodiments, as a preferred implementation, in step S2, the specific formula for constructing the time-varying shear index is as follows:

[0058] In the above formula, α 0 is the base coefficient; c 1~ c 4 represents the coefficients used to fit the historical meteorological tower data; S ( t The stability is categorized into levels, with -1 for unstable, 0 for neutral, and +1 for stable; PBLH is the planetary boundary layer height. The stability categorization is based on the following criteria: strong solar radiation during the day and a thick PBLH indicate unstable conditions; radiative cooling at night and a shallow PBLH indicate stable conditions; and all other conditions are classified as neutral. Roughness is greater in spring when vegetation is abundant and decreases in autumn and winter when vegetation withers.

[0059] Based on the dynamic roughness length and topographic information, a physically consistent baseline wind speed needs to be established as a reference for subsequent corrections. Since reanalysis data typically lack parameters such as friction velocity, this embodiment employs a power-law extrapolation method with a 10m wind speed as the anchor point, and combines stability and surface features to construct a time-varying shear index.

[0060] Further analysis of hourly 10m wind speed and direction, planetary boundary layer height (PBLH), near-Earth meteorological elements (temperature, humidity, cloud cover, incident shortwave, etc.), and dynamic roughness length. z 0( t ) and seasonal vegetation parameters (NDVI / LAI).

[0061] Stability classification: Based on PBLH, cloud cover, and day / night information, atmospheric stability is classified into three categories: unstable (-1), neutral (0), and stable (+1). For example, strong solar radiation and a thick boundary layer during the day are classified as unstable; radiative cooling and a shallow boundary layer at night are classified as stable; and other conditions are classified as neutral.

[0062] Shear index construction: Based on a height of 10m as the anchor point, stability grading, vegetation index, and roughness are introduced to construct a time-varying shear index. α ( t ).

[0063] Power-law extrapolation, using the time-varying shear index, extrapolates a 10m wind speed to the target height:

[0064] Wherein, U10 represents the reanalysis 10m wind speed. α ( t The shear index is a time-varying index. Through this process, the baseline wind speeds at 100m and 150m can be obtained as a function of time. These wind speeds reflect the anchoring information of the reanalysis data and incorporate the effects of stability and surface roughness, thus ensuring the physical consistency of the results and providing a reliable reference for subsequent residual correction.

[0065] Based on the above embodiments, as a preferred implementation, in step S3, the hourly feature vector includes wind speed difference, wind direction sine and cosine coding, terrain upwind index, stability classification, PBLH and cloud cover parameters in the reanalysis data, and data quality markers; wherein, the wind speed difference is the wind speed difference between the target height and the near-ground reference height; the terrain upwind index is the difference between wind direction and slope aspect; and the data quality markers include complement markers and low-trust markers.

[0066] Specifically, in this embodiment, after obtaining the physical baseline wind speed, it is necessary to construct the input features for the machine learning model. The goal of this step is to organize the previously obtained observation, reanalysis, terrain, and baseline information into a unified hourly feature vector, ensuring rich input dimensions and strict time alignment.

[0067] The specific process is as follows: Input data: wind speed, temperature, humidity, air pressure, reanalysis data (10m / 100m wind speed, PBLH, cloud cover, shortwave radiation, etc.) observed at the tower station, topographic features (slope, aspect, roughness), and baseline wind speed results.

[0068] Processing procedure: Time alignment: Aggregate 10-minute data from the tower station to the hour, synchronizing with hourly data from the reanalysis.

[0069] Spatial registration: The grid data was reanalyzed and bilinearly interpolated to the latitude and longitude of the tower station.

[0070] Feature construction includes wind direction sine and cosine coding, wind speed difference (such as the difference between 100m and 10m), topographic windward index (difference between wind direction and slope aspect), stability category (inferred from PBLH and near-surface temperature and humidity), and tower-baseline difference.

[0071] Quality information: The complement value and confidence level are included in the feature.

[0072] Output: Hourly feature vector Xt is generated, which strictly corresponds to the label (the difference between the observed wind speed and the baseline wind speed, i.e., the residual).

[0073] This step ensures the integrity and consistency of the data and provides multi-dimensional inputs from physical, statistical, and environmental perspectives for machine learning models.

[0074] After obtaining features and labels, the core residual learning stage begins. The idea here is to let the machine learning model learn the measured wind speed at the tower. The residual portion of the baseline wind speed is used to correct systematic biases in the reanalysis data.

[0075] The specific process is as follows: Input data: hourly feature vector X; and residual labels y = U obs - U base .

[0076] Processing procedure: 1. Model selection: Random forest (RF) or quantile random forest (QRF) can be used to fit nonlinear relationships and output uncertainty distribution.

[0077] 2. Training strategy: Use hierarchical cross-validation with one station left out and one season left out to ensure that the model has generalization ability in both space and season.

[0078] 3. Sample weights: Reduce the weights of low-confidence samples (such as those that are frozen, stopped, or have been supplemented) to avoid misleading the model.

[0079] 4. Feature screening: Redundant or invalid features are eliminated through dual checks of permutation importance and out-of-bag error.

[0080] Output result: Point prediction residual The corresponding corrected wind speed Û = U base + : Quantile predictions (such as 5%, 50%, 95%) provide support for subsequent interval assessments. Training reports include accuracy metrics such as MAE, RMSE, and R².

[0081] This step combines physics and machine learning, entrusting the complex biases that are difficult to model directly to statistical models.

[0082] Based on the above embodiments, as a preferred implementation, in step S4, when the wind speed residual of the target wind resource assessment area is predicted according to the supervised prediction model, if it is determined that the wind speed data of the multi-height-level wind measurement towers in the target wind resource assessment area has less than 3 months of continuous observation time, a source domain subgroup with a similarity of ≥80% to the climate and terrain features of the target domain is selected from the accumulated multiple complete data stations. The kernel mean matching (KMM) method is adopted, and the hourly feature vector constructed in S3 is used as the feature dimension to adjust the sample weights of the source domain subgroup, so that the Wasserstein distance between the adjusted source domain feature distribution and the feature distribution of a small number of samples in the target wind resource assessment area is ≤0.2, thus eliminating the distribution offset. The tree structure parameters of the supervised prediction model trained by S3 are frozen. Based on the multi-height wind measurement tower data of the target wind resource assessment area, the output bias of the leaf nodes of the supervised prediction model is fine-tuned with L2 regularization. The regularization coefficient is... λ =0.01, so that the residual prediction MAE of the supervised prediction model in the target domain with few samples after fine-tuning is reduced by ≥15% compared with that before fine-tuning. After completing the cross-site migration adaptation, the wind speed residual prediction of the target wind resource assessment area is then carried out.

[0083] Specifically, after the model training is completed, the correction method can be applied to target areas with no or few towers, so as to realize the rapid implementation of wind resource correction results.

[0084] Input data: hourly reanalysis data of the target point (10m wind speed, stability, PBLH, etc.), DEM topographic information and dynamic roughness, and baseline wind speed calculation parameters.

[0085] The baseline wind speed at the target point is calculated, and hourly feature vectors are constructed (relying only on available reanalysis and terrain features in the case of no wind tower). These features are then input into a supervised prediction model to obtain residual predictions and quantiles. The residuals are added to the baseline to obtain the final corrected wind speed. This results in an hourly corrected wind speed sequence and corresponding interval estimation results. This process ensures that the method can be applied not only to sites with wind towers but also to new sites without towers, providing support for wind power planning.

[0086] Based on the above embodiments, as a preferred implementation, in step S4, physical calibration is further performed through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limiting, including: An equidistant regression algorithm is used, with the initial corrected wind speeds at heights of 100m and 150m as inputs. If the initial corrected wind speed at 150m is less than the initial corrected wind speed at 100m, or if the wind speed at either height is less than 0m / s, the wind speeds at both heights are adjusted synchronously through equidistant regression. After adjustment, the final corrected wind speed at 150m must be greater than or equal to the final corrected wind speed at 100m, and the wind speeds at both heights must be greater than or equal to 0m / s. The absolute deviation between the adjusted wind speed and the initial corrected wind speed must be less than or equal to 0.5m / s. Based on the terrain roughness in the terrain and surface feature data, the roughness difference between adjacent grid points is calculated. If the roughness difference exceeds the preset range threshold, the corrected wind speed in the roughness transition area is smoothed by Gaussian filtering. Wind speed upper limit thresholds are set according to different climate zones and stability conditions. When the predicted wind speed exceeds the upper limit threshold, it is truncated and marked. The mark needs to be output along with the final corrected wind speed.

[0087] The following are examples of the methods described in embodiments of the present invention. Example 1 (Mountain wind field, 100m level) Data and region: A mountain wind field, tower 70 / 100 / 150m (10min), 2019–2023; ERA5-Land (1h); DEM 30m; LC and NDVI / LAI monthly scale.

[0088] Process: Implemented according to the above technical solution, QRF is used for residual learning, and the wind speed correction results are generated through post-calibration and uncertainty assessment.

[0089] Results: The annual MAE decreased from 1.92 m / s to 1.15 m / s, R² increased to 0.935, and PICP (90%) increased to 90% after conformal calibration.

[0090] Example 2 (coastal wind field, 150m layer, few sample migrations) Data and region: New coastal station, only 2 months of tower 100 / 150m data; ERA5-Land; DEM, LC, NDVI.

[0091] Migration strategy: Climate gating is differentiated by KöppenCfa / Cwa; KMM reweighting; forest structure is frozen, and L2 regularization is only applied to leaf node bias.

[0092] Results: MAE decreased from 2.10 m / s to 1.31 m / s, R² increased from 0.79 to 0.91, and PICP (80%) improved to 81% after conformal calibration.

[0093] Optional Implementation (Satellite Stability Proxy) By introducing FY-4B cloud parameters / shortwave radiation components as stability surrogate features, the model's perception of stability is enhanced when the reanalysis boundary layer index is unreliable, thereby improving the correction robustness under nighttime and cloudy conditions.

[0094] Secondly, an embodiment of the present invention provides a wind speed deviation correction system based on the fusion of wind tower data and reanalysis data, and a wind speed deviation correction method based on the fusion of wind tower data and reanalysis data in the above-described format example, such as... Figure 2 As shown, it includes: The data acquisition and preprocessing module 210 acquires multi-height-level wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and performs spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. The baseline wind speed construction module 220 uses the near-Earth reference height wind speed from the reanalysis data as an anchor point to obtain dynamic roughness. Based on the stability classification of planetary boundary layer height (PBLH), cloud cover, and day / night information, the dynamic roughness and stability classification are integrated into the construction process of the time-varying shear index to obtain the time-varying shear index that changes dynamically over time. Based on the power-law extrapolation formula, the near-Earth reference height wind speed is extrapolated to the target height using the time-varying shear index as the core parameter to obtain the physically consistent baseline wind speed. The feature integration and model training module 230 integrates the standardized dataset and the physically consistent baseline wind speed into hourly feature vectors. The difference between the wind speed measured by the wind tower at the same time and the baseline wind speed is used as the residual label. A sample set with one-to-one correspondence between the hourly feature vectors and the residual labels of the same period is constructed. A supervised prediction model that can predict wind speed residuals and uncertainties is trained based on the sample set. The residual prediction and physical calibration module 240 predicts the wind speed residual of the target wind resource assessment area according to the supervised prediction model, superimposes the wind speed residual with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed, and then performs physical calibration through high monotonicity constraints, roughness consistency smoothing and extreme wind speed limits to obtain the final corrected wind speed and the corresponding uncertainty confidence interval.

[0095] Based on the same concept, this invention also provides a schematic diagram of a physical structure, such as... Figure 3As shown, the server may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions stored in the memory 330 to execute the steps of the wind speed deviation correction method for fusing wind tower data and reanalysis data as described in the above embodiments.

[0096] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] Based on the same concept, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program containing at least one piece of code that can be executed by a master control device to control the master control device to implement the steps of the wind speed deviation correction method for fusing wind tower data and reanalysis data as described in the above embodiments.

[0098] Based on the same technical concept, this application also provides a computer program, which, when executed by a main control device, is used to implement the above-described method embodiments.

[0099] The program may be stored, in whole or in part, on a storage medium packaged with the processor, or in part or in whole on a memory not packaged with the processor.

[0100] Based on the same technical concept, this application also provides a processor for implementing the above-described method embodiments. The processor can be a chip.

[0101] The various embodiments of the present invention can be combined arbitrarily to achieve different technical effects.

[0102] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive).

[0103] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0104] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for correcting wind speed deviations by fusing anemometer data and reanalysis data, characterized in that, include: S1. Obtain multi-height wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and perform spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. S2. Using the near-Earth reference height wind speed of the reanalysis data as the anchor point, dynamic roughness is obtained. Based on the stability classification of planetary boundary layer height (PBLH), cloud cover, and day and night information, dynamic roughness and stability classification are integrated into the construction process of time-varying shear index to obtain time-varying shear index that changes dynamically with time. Based on the power-law extrapolation formula, using the time-varying shear index as the core parameter, the wind speed at the near-ground reference height is extrapolated to the target height to obtain the physically consistent baseline wind speed. S3. Integrate the standardized dataset and the physical consistency baseline wind speed into hourly feature vectors. Use the difference between the wind speed measured by the wind tower at the same time and the baseline wind speed as the residual label. Construct a sample set in which the hourly feature vectors and the residual labels of the same period correspond one-to-one. Train a supervised prediction model that can predict wind speed residuals and uncertainties based on the sample set. S4. Based on the wind speed residuals predicted by the supervised prediction model, the wind speed residuals are superimposed with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed. Then, physical calibration is performed through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limiting to obtain the final corrected wind speed and the corresponding uncertainty confidence interval. In step S2, obtaining the dynamic roughness includes: Calculate terrain roughness based on the Digital Elevation Model (DEM): In the above formula, R is the terrain roughness index, in meters (m). z i For the first DEM window i Elevation values ​​of each grid point; The average elevation of the DEM window; n This represents the number of grid points within the DEM window. Dynamic roughness is calculated by combining the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) using an empirical function. In the above formula, z 0,terrain The basic terrain roughness data obtained from DEM calculation, a , b These are empirical coefficients fitted to historical wind measurement tower data. t It is a time variable; NDVI ( t )for t Normalized Difference Vegetation Index (LAI) at any given time t )for t Leaf area index at time; The specific formula for constructing the time-varying shear index is as follows: In the above formula, α 0 is the base coefficient; c 1~ c 4 represents the coefficients used to fit the historical meteorological tower data; S ( t The stability is categorized into levels, with -1 for unstable, 0 for neutral, and +1 for stable; PBLH is the planetary boundary layer height; the stability categorization is based on the following criteria: strong solar radiation during the day and a thick PBLH indicate unstable conditions, radiative cooling at night and a shallow PBLH indicate stable conditions, and all other conditions are classified as neutral.

2. The wind speed deviation correction method based on the fusion of wind tower data and reanalysis data according to claim 1, characterized in that, In S1, the multi-height-layer wind measurement tower data includes wind speed data with a continuous observation period of at least one year, the resolution of the wind speed data is 10 minutes, the wind speed data covers at least two hundred-meter-level heights of 100m and 150m, and the multi-height-layer wind measurement tower data also includes temperature, humidity, and air pressure. The reanalysis data includes wind speed, air temperature, air pressure, humidity, friction speed, stability, and cloud cover; The terrain and surface feature data include digital elevation model (DEM) data, and terrain roughness data, slope data, and aspect data derived from the DEM data, as well as seasonal normalized vegetation index (NDVI) data and leaf area index (LAI) data. Step S1 also includes acquiring satellite data of the target wind resource assessment area, including cloud cover and shortwave radiation parameters.

3. The wind speed deviation correction method based on the fusion of wind tower data and reanalysis data according to claim 1, characterized in that, In S3, the hourly feature vector includes wind speed difference, wind direction sine and cosine coding, terrain upwind index, stability classification, PBLH and cloud cover parameters in the reanalysis data, and data quality markers; wherein, the wind speed difference is the wind speed difference between the target height and the near-ground reference height; the terrain upwind index is the difference between wind direction and slope aspect; and the data quality markers include complement markers and low-trust markers.

4. The wind speed deviation correction method based on the fusion of wind tower data and reanalysis data according to claim 1, characterized in that, In S4, when the wind speed residual of the target wind resource assessment area is predicted according to the supervised prediction model, if it is determined that the wind speed data of the multi-height-level wind measurement towers in the target wind resource assessment area has less than 3 months of continuous observation time, the source domain subgroup with a similarity of ≥80% to the climate and terrain features of the target domain is selected from the accumulated multiple complete data stations. The kernel mean matching (KMM) method is adopted, and the hourly feature vector constructed in S3 is used as the feature dimension to adjust the sample weights of the source domain subgroup, so that the Wasserstein distance between the adjusted source domain feature distribution and the feature distribution of a small number of samples in the target wind resource assessment area is ≤0.2, thus eliminating the distribution offset. The tree structure parameters of the supervised prediction model trained by S3 are frozen. Based on the multi-height wind measurement tower data of the target wind resource assessment area, the output bias of the leaf nodes of the supervised prediction model is fine-tuned with L2 regularization. The regularization coefficient is... λ =0.01, so that the residual prediction MAE of the supervised prediction model in the target domain with few samples after fine-tuning is reduced by ≥15% compared with that before fine-tuning. After completing the cross-site migration adaptation, the wind speed residual prediction of the target wind resource assessment area is then carried out.

5. The wind speed deviation correction method based on the fusion of wind tower data and reanalysis data according to claim 4, characterized in that, In step S4, physical calibration is performed through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limiting, including: An equidistant regression algorithm is used, with the initial corrected wind speeds at heights of 100m and 150m as inputs. If the initial corrected wind speed at 150m is less than the initial corrected wind speed at 100m, or if the wind speed at either height is less than 0m / s, the wind speeds at the two heights are adjusted synchronously through equidistant regression. After adjustment, the final corrected wind speed at 150m must be greater than or equal to the final corrected wind speed at 100m, and the wind speeds at both heights must be greater than or equal to 0m / s. The absolute deviation between the adjusted wind speed and the initial corrected wind speed must be less than or equal to 0.5m / s. Based on the terrain roughness in the terrain and surface feature data, the roughness difference between adjacent grid points is calculated. If the roughness difference exceeds the preset range threshold, the corrected wind speed in the roughness transition area is smoothed by Gaussian filtering. Wind speed upper limit thresholds are set according to different climate zones and stability conditions. When the predicted wind speed exceeds the upper limit threshold, it is truncated and marked. The mark needs to be output along with the final corrected wind speed.

6. A wind speed deviation correction system for fusing anemometer data and reanalysis data, used to execute the wind speed deviation correction method for fusing anemometer data and reanalysis data as described in any one of claims 1 to 5, characterized in that, include: The data acquisition and preprocessing module acquires multi-height-level wind measurement tower data, reanalysis data, and topographic and surface feature data of the target wind resource assessment area, and performs spatiotemporal matching, outlier removal, missing value filling, and consistency verification to obtain a standardized dataset with spatiotemporal matching. The baseline wind speed construction module uses the near-Earth reference height wind speed of the reanalysis data as the anchor point to obtain dynamic roughness. Based on the stability classification of planetary boundary layer height (PBLH), cloud cover, and day and night information, the dynamic roughness and stability classification are integrated into the construction process of the time-varying shear index to obtain the time-varying shear index that changes dynamically over time. Based on the power-law extrapolation formula, using the time-varying shear index as the core parameter, the wind speed at the near-ground reference height is extrapolated to the target height to obtain the physically consistent baseline wind speed. The feature integration and model training module integrates the standardized dataset and the physically consistent baseline wind speed into hourly feature vectors. The difference between the wind speed measured by the wind tower at the same time and the baseline wind speed is used as the residual label. A sample set with one-to-one correspondence between the hourly feature vectors and the residual labels of the same period is constructed. A supervised prediction model that can predict wind speed residuals and uncertainties is trained based on the sample set. The residual prediction and physical calibration module predicts the wind speed residual of the target wind resource assessment area based on the supervised prediction model. It then superimposes the wind speed residual with the physical consistency baseline wind speed to obtain the preliminary corrected wind speed. Finally, it performs physical calibration through high monotonicity constraints, roughness consistency smoothing, and extreme wind speed limits to obtain the final corrected wind speed and the corresponding uncertainty confidence interval.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the wind speed deviation correction method that fuses wind tower data and reanalysis data as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the wind speed deviation correction method that fuses wind tower data and reanalysis data as described in any one of claims 1 to 5.

Images