A variable resolution three-dimensional wind field fusion method for low altitude flight

Abstract

The application discloses a variable-resolution three-dimensional wind field fusion method for low-altitude flight. The method comprises the following steps: firstly, dividing a flight airspace range into three spatial levels of points, lines and surfaces; secondly, acquiring multi-source wind field observation data in the flight airspace range; thirdly, for the surface level, adopting an optimal interpolation method to generate a global basic wind field base map by taking a numerical prediction analysis field of a location of the flight airspace range as a background field and fusing the multi-source wind field observation data; fourthly, for the line level and the point level, adopting a three-dimensional Kriging method to generate a fine-resolution three-dimensional wind field enhancement field, and based on an adaptive selection strategy of a variogram model, adapting the three-dimensional Kriging method to different wind field structure characteristics; and finally, performing spatial transition fusion on the global basic wind field base map and the fine-resolution three-dimensional wind field enhancement field to obtain a spatial variable-resolution three-dimensional wind field. The method can realize rapid generation of high-precision three-dimensional wind fields in different scale regions and under different observation data densities.

Description

Technical Field

[0001] This invention relates to a three-dimensional wind field fusion method and system based on the combination of optimal interpolation (OI) and three-dimensional kriging, which is adapted to the spatial hierarchy of "point-line-surface" for low-altitude flight and has variable resolution. It belongs to the field of low-altitude flight support technology. Background Technology

[0002] The safety and stability of low-altitude flight activities heavily rely on high-precision three-dimensional wind field data. Currently, acquiring three-dimensional wind field data primarily depends on the fusion processing of multi-source meteorological observation data (laser wind radar, wind profiler radar, automatic weather stations, aircraft reports, etc.). Existing fusion technologies mainly include data assimilation methods (3D-Var, 4D-Var) and statistical interpolation methods (Optimal Interpolation (OI), Kriging interpolation). However, existing technologies still have the following shortcomings: First, the globally unified resolution fusion method has technical limitations. This type of method uses a single fusion algorithm and fixed resolution parameters to process the entire target area, failing to simultaneously meet the dual requirements of "global coverage" and "local high precision." For large areas such as entire provinces, using high-resolution parameters results in an exponential increase in the number of grid points and a surge in computational load, making it impossible to meet the real-time requirements of minute-level updates for low-altitude flight. For smaller areas such as cities and airports, using low-resolution parameters leads to the loss of subtle wind field characteristics such as local wind shear and local circulation, failing to provide high-precision support for aircraft attitude control.

[0003] Second, local fixed-resolution fusion methods have technical limitations. These methods employ a single local optimization strategy across the entire target area without dynamically adapting to the density of observation data. Physically constrained methods (such as 4D-Var) cannot fully utilize the statistical advantages of dense data in densely observed areas, resulting in wasted computational resources; statistical interpolation methods (such as Kriging) suffer from significantly increased fusion errors in sparsely observed areas due to a lack of data support, and are prone to producing false wind field information that does not conform to atmospheric motion patterns.

[0004] Third, the integration of global and local wind fields is poor. In existing methods, the global base wind field and the local optimized wind field are often simply superimposed or directly replaced without adaptive smoothing for resolution differences and data deviations. This leads to abrupt changes in the wind field and data discontinuity at the integration point, disrupting the spatiotemporal continuity of the wind field and directly increasing flight safety risks.

[0005] Fourth, the adaptability to low-altitude flight scenarios is insufficient. Aircraft only require ultra-high precision wind fields in key areas such as takeoff and landing points, flight path turning points, and complex terrain crossing points, while other areas only require basic precision. Existing methods do not construct a variable resolution fusion architecture that allocates resources on demand, resulting in wasted computing resources in non-critical areas and insufficient precision in critical areas, making it difficult to adapt to the diverse and refined needs of low-altitude flight.

[0006] In summary, existing technologies cannot efficiently generate variable-resolution 3D wind fields suitable for low-altitude flight requirements. Therefore, there is an urgent need to develop a fusion method that can dynamically adapt the fusion strategy for different regional scales and different observation data densities, and can achieve seamless integration of the global base field and the local enhanced field. Summary of the Invention

[0007] The technical problem this invention aims to solve is: the contradiction between single-resolution wind field and large-scale coverage and detailed characterization of key areas; the problem of discontinuity and physical inconsistency in sparse observation areas; and the differentiated accuracy requirements of low-altitude flight missions for different spatial locations (take-off / landing points / flight routes / areas).

[0008] This invention aims to address the technical problems of existing 3D wind field fusion methods, such as the inability to simultaneously achieve global coverage and high local precision, poor adaptability of observation data density, and unnatural fusion transitions. To this end, it provides a variable-resolution 3D wind field fusion method for low-altitude flight. This method is applicable to scenarios such as flight path planning, flight safety early warning, and flight attitude control for low-altitude aircraft (e.g., helicopters, light aircraft, UAVs), and can achieve rapid generation of high-precision 3D wind fields under different scale regions and different observation data densities.

[0009] This invention is achieved using the following technical solution: A variable-resolution 3D wind field fusion method for low-altitude flight includes the following steps: 1. Obtain low-altitude flight missions and classify spatial levels. Obtain the coordinates of the take-off and landing points, flight path, and airspace range in the flight mission; based on the flight mission, divide the airspace range into point-level areas, line-level areas, and area-level areas according to three spatial levels: point level, line level, and area level. The point-level area is the area surrounding the take-off and landing points and special terrain points (such as high mountains, canyons, high-altitude areas, etc., which may affect flight safety). The line-level area is the buffer zone on both sides of the flight path. The area-level area is the entire airspace range.

[0010] 2. Acquire multi-source wind field observation data and perform quality control. Acquire multi-source wind field observation data within the flight airspace, including wind field inversion data from lidar and weather radar, as well as wind field data monitored by wind profiler radar, automatic ground stations, and aircraft reports (including wind profiler wind field, automatic ground station wind field observation data, urban canopy ultrasonic wind field, wind tower gradient wind field, and aircraft report wind field data); and perform quality control on the multi-source wind field observation data (such as outlier removal and defuzzification).

[0011] 3. Generate a global base wind field map The appropriate fusion algorithm is selected based on the scale of the flight airspace to generate a global base wind field map. The specific method is as follows: Using the optimal interpolation algorithm, the numerical forecast analysis field of the flight airspace is used as the background field. Multi-source wind field observation data with quality control are fused to generate a basic resolution three-dimensional wind field base map, which is the global basic wind field base map (hereinafter referred to as the global basic field). The core formula of the optimal interpolation algorithm is:

[0012] In the formula, and These represent the analyzed value and the initial estimate at grid point g of the wind field grid, respectively. and These represent the wind field observation and the initial estimate of the background field at grid point i near grid point g, respectively. represents the weighting coefficient when estimating the bias of grid point i in the wind field grid, and N is the total number of grid points when estimating the bias of grid point i in the wind field grid.

[0013] It should be noted that the global base wind field map covers all grids in the area where the flight airspace is located (related to the flight scale; for example, if the flight scale is a province, then the entire province needs to be covered), including the line-level and point-level areas that will be covered by the enhanced field later, in order to ensure the continuity and physical consistency of the entire field data and provide complete base map data for subsequent transition fusion.

[0014] 4. Local enhancement of key flight areas. Point-level and line-level regions are extracted as key flight regions. A three-dimensional kriging algorithm is used to generate a fine-resolution three-dimensional wind field enhancement field (referred to as a local enhancement field) corresponding to the line-level or point-level regions without introducing a background field.

[0015] The core formula of the 3D Kriging algorithm is:

[0016] in, These are the interpolation points (target wind field grid points).g The estimated wind field value, For the first Wind field observations at each wind field grid point n To determine the total number of wind field grid points within the study area, For the first The interpolation weights for each observation point, to ensure unbiasedness, satisfy the following conditions: The optimal weights are obtained by minimizing the estimated variance, which ultimately boils down to solving the Kriging equations:

[0017] In the formula, Let be the semivariogram (or simply variogram), representing the _th _ i The and the first j A measure of wind field variability among individual wind field grid points; Indicates the first i The and the first g A measure of wind field variability among individual wind field grid points; These are Lagrange multipliers. In three-dimensional space, the variogram is usually assumed to be isotropic, meaning it depends only on the Euclidean distance between two points. .

[0018] The variogram is the core of the Kriging method. It reflects the structural characteristics of spatial variables. C0 is the nugget constant, reflecting microvariation and measurement error; C is the sill value, reflecting structural variability; and a is the range, representing the maximum distance of spatial correlation. Commonly used variogram models include: The spherical model is the most commonly used variogram model. It exhibits linear growth near the origin and stabilizes after reaching a range 'a'. It is suitable for wind field structures where spatial correlation linearly decreases with distance and then stabilizes, such as wind fields in stratiform cloud systems, the outer regions of large-scale weather systems (e.g., fronts, cyclones), and near-surface wind fields in flat terrain. Its function is specifically:

[0019] The exponential model has a steep slope near the origin, and at an effective range of 3a, the variogram value reaches 95% of the base value, indicating it tends to be stable. It is suitable for wind field structures where spatial correlation decays exponentially with distance, such as small- to medium-scale convective systems (e.g., squall lines, convective cells), local wind shear regions, and wind fields in urban built-up areas (local turbulence caused by buildings). Its function is specifically:

[0020] The Gaussian model has a zero slope and slow growth near the origin, and its effective range is... At a point where the variogram value reaches 95% of the base value, it can be considered to be stable. This is applicable to wind field structures with extremely strong spatial continuity and very smooth changes, such as the eye of a typhoon and its outer spiral rainbands, wind fields under large-scale stable stratification, and wind fields over water surfaces (oceans, lakes), etc. Its specific function is as follows:

[0021] Theoretically, spherical models are suitable for wind fields with large-scale uniform variations, exponential models are suitable for wind fields with drastic local variations, and Gaussian models are suitable for extremely smooth and highly continuous wind fields. However, critical flight areas (such as flight paths and takeoff and landing fields) may simultaneously contain multiple wind field structural features, and the spatial structure of the wind field in the same area also changes dynamically with the evolution of the weather system. If a single fixed model is used, it is difficult to adapt to such complex and variable realities, and the interpolation accuracy will be significantly affected. Therefore, this preferred scheme innovatively proposes an adaptive selection strategy for the variogram model, the steps of which are as follows: Step 1: Calculate the experimental variation function. Assume there are N observation points near the current target wind field grid point, and the distance between any two points i and j is... The corresponding wind field observation values ​​are as follows: and , distance According to step size Grouping, for the k-th group, we have A pair of points (any two points form a pair of points) The midpoint distance of this group (e.g., if grouped by 100 meters, the first group is 0-100, then the midpoint distance of this group is (0+100) / 2=50, and so on), and its experimental variogram value is:

[0022] The second step involves performing weighted least-squares fitting of the experimental variation function using a spherical model, an exponential model, and a Gaussian model, respectively. The optimal parameters are then determined by solving the following optimization problem. Where K is the total number of lag distance groups, Weighting coefficients (usually taken as...) (i.e., the number of point pairs in the k-th group). Let be the experimental variation function value for the k-th group. For theoretical models (including spherical models, exponential models, and Gaussian models) in Value:

[0023] The third step is to calculate the sum of squared residuals (RSS) for each model:

[0024]

[0025] in Indicates the optimal parameters Down, The corresponding variation function value of the model.

[0026] The fourth step is to select the model with the smallest sum of squared residuals (RSS) as the variation function model for the current region.

[0027] This strategy calculates the goodness of fit of spherical, exponential, and Gaussian models based on observation data near the current target wind field grid points, and automatically selects the optimal model. If the RSS values ​​of multiple models are similar (e.g., relative differences less than 5%), the spherical model with fewer parameters and clear physical meaning is preferred. This allows Kriging interpolation to dynamically adapt to different wind field structures, thereby improving the accuracy and adaptability of wind field enhancement in key areas.

[0028] 5. Spatial transition and integration The global base wind field map (surface level) generated in step 3 is then fused and stitched together with the fine-resolution 3D wind field enhancement field (point level and line level) generated in step 4: The first step is to perform data preprocessing on the fine-resolution 3D wind field enhancement field, including aligning the time coordinates of the global base wind field map and the fine-resolution 3D wind field enhancement field to the same moment, projecting the fine-resolution 3D wind field enhancement field grid onto the output grid coordinate system through bilinear interpolation to achieve spatial registration, and removing outliers, to ensure spatiotemporal consistency and reliable data quality.

[0029] The second step involves fusing the overlapping regions of the surface layer and the point layer, as well as the surface layer and the line layer, using an adaptive weighting function. The fusing formula is as follows:

[0030] in, This represents the coordinates of the wind field grid points. To integrate the wind field grid points The wind field value, Enhanced field grid points for high-resolution 3D wind field The wind field value, To enhance the field weight of the three-dimensional wind field with high resolution, Grid points for the global base wind field map The wind field value, The weights are assigned to the global base wind field map. The weight allocation formula is based on the distance quantization between the grid points to be fused and the center of the fine-resolution 3D wind field enhancement field, as follows:

[0031]

[0032] in d The distance (in km) between the grid points of the grid to be fused and the center of the fine-resolution 3D wind field enhancement field. d 0 represents the radius of the enhanced three-dimensional wind field region with high resolution. d 1 represents the distance between the outer boundary of the overlapping region and the center of the fine-resolution 3D wind field enhancement field. That is, the weight of the center region of the enhancement field is set to 75%. At the edge of the overlapping region, the weight of the enhancement field decreases linearly from 75% to 25%, while the weight of the global basic wind field map increases linearly from 25% to 75%. This can achieve a gradual transition between the global basic wind field map and the fine-resolution 3D wind field enhancement field, avoiding sudden changes in the wind field.

[0033] The third step, to prevent minor discontinuities at the fusion boundary, involves Gaussian smoothing of the fusion boundary (edge ​​region) between the global base wind field map and the enhanced field:

[0034] in, Smoothed wind field grid points The wind field value, The unit is grid points (based on pilot data from Zhejiang Province after calibration). Take 0.8). The radius of the Gaussian kernel is 1. Represents the grid points of the merged wind field. The wind field value; i and j are the number of grid points moved in the x and y directions, respectively; , The grid spacings are for the x and y directions, respectively. This formula eliminates abrupt changes in the wind field at the edges, ensuring the spatiotemporal continuity of the wind field while preserving subtle wind field details such as local wind shear and gusts.

[0035] The final output is a spatially variable resolution three-dimensional wind field. This wind field maintains a basic resolution (0.01°×0.01°×200m, time resolution of 10 minutes) at the surface level, and achieves higher resolution (0.001°×0.001°×50m, time resolution of 3 minutes) at the point and line levels. Moreover, the resolution smoothly transitions to the basic resolution as the distance from the flight path or take-off and landing point increases, ensuring the spatiotemporal continuity of the wind field and meeting the real-time requirements of minute-level updates for low-altitude flight.

[0036] This invention also provides a variable-resolution three-dimensional wind field fusion system for low-altitude flight, which is used to execute the aforementioned variable-resolution three-dimensional wind field fusion method for low-altitude flight. The system includes: The spatial hierarchy division module is used to obtain the take-off and landing point coordinates, flight path and flight airspace range of low-altitude flight missions; and based on the low-altitude flight missions, divide the flight airspace range into three spatial levels: point level, line level and area level. The data acquisition module acquires multi-source wind field observation data within the flight airspace and performs quality control. The global basic wind field base map generation module is used to generate a global basic wind field base map for the surface level by using the optimal interpolation method, taking the numerical forecast analysis field of the flight airspace as the background field, and fusing quality-controlled multi-source wind field observation data. The fine-resolution 3D wind field enhancement field generation module is used to generate fine-resolution 3D wind field enhancement fields corresponding to the point and line levels using the 3D Kriging method. Based on the adaptive selection strategy of the variation function model, the 3D Kriging method is adapted to different wind field structure characteristics. The spatial transition fusion module is used to perform spatial transition fusion between the global basic wind field base map and the fine resolution three-dimensional wind field enhancement field. The fusion is performed in the overlapping areas of the surface layer and the point layer, as well as the surface layer and the line layer. The edge areas of the surface layer and the point layer, as well as the surface layer and the line layer, are processed by the Gaussian smoothing algorithm to obtain a spatially variable resolution three-dimensional wind field.

[0037] The present invention also provides an electronic device comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the variable resolution three-dimensional wind field fusion method for low-altitude flight.

[0038] The present invention also provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to perform the steps of the variable resolution three-dimensional wind field fusion method for low-altitude flight.

[0039] The beneficial effects of this invention are as follows: 1. Strong regional scale adaptability: The fusion algorithm can be dynamically adapted according to the flight area scale, realizing "on-demand precision" variable resolution fusion, taking into account the global coverage and physical coordination of wind fields in large areas, as well as the high precision requirements of small areas, and adapting to low-altitude flight scenarios of different scales.

[0040] 2. It solves the fundamental defect of discontinuity of optimal interpolation in sparse observation regions: instead of trying to cover all scenes with a single method, it switches to 3D kriging in key areas and uses only high-density observations for pure statistical interpolation, completely avoiding the dependence of OI on the background field and the discontinuity of local analysis.

[0041] 3. Significantly improves the accuracy and adaptability of wind field enhancement in key areas: This invention introduces a "fixed method, adaptive parameters" strategy in three-dimensional kriging interpolation, enabling the algorithm to automatically optimize internal parameters based on local environmental conditions and observation data characteristics, significantly improving the accuracy of wind field enhancement in key areas under complex terrain and different weather conditions.

[0042] 4. Seamless integration of global base field and local enhancement field: This invention constructs a smooth transition mechanism based on spatial distance between the global base field and the local enhancement field. By designing adaptive weighting functions for point and line levels, the weighting coefficients dynamically change with spatial location. After fusion, Gaussian smoothing is applied to the transition zone, and possible minor discontinuities are eliminated through neighborhood weighted averaging, making the wind field more spatially coordinated and natural.

[0043] 5. High computational efficiency, meeting the real-time requirements of low-altitude flight: Variable resolution architecture design and algorithm optimization significantly reduce computational load, with time lag controlled within 3 minutes, meeting the real-time support needs of low-altitude flight. Wind field parameters are compatible with low-altitude flight-related systems, requiring no additional adaptation and applicable to various low-altitude flight scenarios, making it highly valuable for widespread adoption. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method proposed in this invention; Figure 2 The relationship between the root mean square error of the background wind field at any two points on the U component and the distance between the two points; Figure 3 The variation characteristics of the vertical direction error correlation coefficient of the U component; Figure 4 The relationship between the root mean square error of the background wind field at any two points on the V component and the distance between the two points; Figure 5 The variation characteristics of the vertical direction error correlation coefficient of the V component. Detailed Implementation

[0045] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Example 1 This invention provides a variable resolution three-dimensional wind field fusion method for low-altitude flight (the process is as follows). Figure 1As shown), the method specifically includes the following steps: First, spatial hierarchy identification and division are performed.

[0047] Obtain the coordinates of the take-off and landing points, flight path, and airspace range in the flight mission; based on the flight mission, divide the airspace range into point-level regions, line-level regions, and area-level regions according to three spatial levels: point level, line level, and area level.

[0048] In this embodiment, the different level recognition parameters are set as follows: point level threshold. Line-level threshold Business research revealed that wind field changes within a 100-500m radius of the takeoff and landing site have the most significant impact on UAV takeoffs and landings, while the 1-2km radius on either side of the flight path is the primary focus area for flight path risk assessment. Considering that UAVs fly at speeds of 5-10m / s during takeoff and landing, a point-level threshold of 1km corresponds to approximately 100-200 seconds of flight time during the UAV takeoff and landing phase, which is sufficient to cover the entire process from takeoff to climb to a safe altitude (or from descent to landing), ensuring high-precision wind field protection during critical takeoff and landing phases. A line-level threshold of 2km corresponds to approximately 100 seconds of lateral deviation warning range during the UAV cruise phase, covering the trajectory deviation range under typical crosswind influences.

[0049] For any grid point G within the target space, the rule for determining its spatial hierarchy is: if ( For any single target point, such as a takeoff or landing point, or a point in a terrain area that could affect flight safety (such as mountains, valleys, or high-altitude areas), it is divided into a point-level region; if ( For any linear target (i.e., flight path), it is divided into a line-level area; the entire flight airspace is a surface-level area.

[0050] Specifically, at the surface level: the entire Zhejiang Province, serving as the basic wind field coverage area, with a horizontal range covering the Zhejiang region from 117.5°E to 123.1°E and from 26.8°N to 31.6°N, and a vertical range from 100 to 2900m.

[0051] Line level: 2km buffer zone on both sides of each air route. Taking the Hangzhou low-altitude medical delivery air route as an example, from the Provincial Blood Center in the city center to the First Affiliated Hospital of Zhejiang University in the west of the city, the entire 18km route crosses half of the main urban area, including multiple route turning points, and the line level covers the surrounding area of ​​the air route.

[0052] Point level: The area within 1km of each take-off and landing site. Taking the Hangzhou Yuhang Low-Altitude Test Zone as an example, it includes several take-off and landing sites; taking Jiaxing Haining as an example, it includes 73 UAV take-off and landing points; taking Taizhou Linhai as an example, it includes UAV operation take-off and landing points in citrus planting areas.

[0053] Then, multi-source wind field observation data are acquired and quality control is performed.

[0054] Wind field data from various sources, including lidar-derived wind fields, weather radar-derived wind fields, wind profiler wind fields, wind field observations from ground-based automatic weather stations, urban canopy ultrasonic wind fields, wind tower gradient wind fields, and aircraft-reported wind field data, were used as the observation fields. For lidar-derived wind fields, gross errors in the corresponding radial velocity were eliminated based on a signal-to-noise ratio threshold, and systematic errors such as wind direction deviation and beam tilting caused by horizontal position were corrected. For X / S band weather radar, an automatic deblurring method was used to deblur the radial velocity, and a dual-radar three-dimensional wind field inversion was performed considering standard atmospheric refraction conditions in the Earth coordinate system. For wind profiler radar, outlier removal was performed using real-time rapid quality control technology based on the principle of temporal and vertical spatial continuity. For wind field observations from ground-based automatic weather stations, urban canopy ultrasonic wind fields, and wind tower gradient wind fields, quality control was performed based on the principles of spatiotemporal consistency and internal consistency. For aircraft-reported wind fields, comparisons were made with numerical weather predictions; data with deviations exceeding 5 m / s were discarded.

[0055] Next, a global base wind field map is generated.

[0056] Using the three-dimensional atmospheric analysis field generated by numerical weather prediction as the background field, with a horizontal resolution of 0.03°×0.03°, a vertical resolution of 200m, and a temporal resolution of 10 minutes, the optimal interpolation method is used to fuse multi-source wind field observation data after quality control to generate a base-resolution three-dimensional wind field map, which is the global base wind field map (referred to as the global base field, 0.01°×0.01°×200m, with a temporal resolution of 10 minutes). The calculation method for the wind field analysis value (U and V components calculated separately) at each grid point g of the wind field is as follows: A weighted estimate is performed on the deviations between the wind field observations (lidar, wind profiler radar, etc.) from the initial estimates of the background field (numerical forecast analysis field) at N wind field grid points within a certain range around wind field grid point g; then, the initial estimate of wind field grid point g is calculated. By adding the weighted estimation results of N wind field grid points, the analytical value at wind field grid point g can be obtained. The specific calculation formula is as follows:

[0057] In the formula, and These represent the wind field observation and the initial estimate of the background field at grid point i near grid point g, respectively. This represents the weighting coefficient used when estimating the deviation of grid point i in the wind field grid. In the optimal interpolation method of this embodiment, the range of deviation calculation is 0.01°×0.01° horizontally and 200m vertically, which is a cuboid range with a side length of 0.02°×0.02°×400m centered on grid point g in the wind field grid, and N is the total number of grid points within this cuboid range.

[0058] Using second-level radiosonde data as the true value T, assume the initial field error is (initial estimate - true value T, the root mean square error of the initial estimate is denoted as...) ), and observation error (observed value - true value T, the root mean square error of observation is denoted as Since the initial field estimation error and the observation error are unbiased, the following linear equation can be constructed using the least squares method. Solving this equation will yield the optimal solution. :

[0059] In the formula, This represents the correlation coefficient between the initial field estimation errors at points i and j. This represents the correlation coefficient of the observation field error between points i and j. This represents the correlation coefficient between the initial field estimation error at point j and point g. Let be the ratio of the observed root mean square error to the initial estimated root mean square error at point i. Let be the ratio of the observed root mean square error to the initial estimated root mean square error at point j.

[0060] The error correlation coefficient is obtained by performing error analysis on each observation data and calculating it using statistical methods. Statistical analysis shows that the error correlation of the background wind field decreases with increasing distance, and a negative exponential function can be used for fitting.

[0061] Figure 2 shows the relationship between the correlation coefficient of the root mean square error of the background wind field at any two points on the U component and the distance between the two points. As shown in the figure, the correlation of the background wind field error in the U component decreases with increasing distance: the horizontal (zonal and meridional) decorrelation distance is 25 km, and the vertical decorrelation distance is 1 km. The yellow solid line and green dashed line in the figure are the negative exponential fitting curves of the zonal and meridional error correlation coefficients, respectively, and the red solid line is the negative exponential fitting curve of the vertical error correlation coefficient. All three types of curves accurately depict the attenuation law in their respective directions. Figure 3 shows the variation characteristics of the vertical error correlation coefficient of the U component. As shown in the figure, the vertical error correlation of the U component decreases rapidly within 1 km, verifying the rationality of the vertical decorrelation distance. Figure 4 shows the relationship between the correlation coefficient of the root mean square error of the background wind field at any two points on the V component and the distance between the two points. As can be seen from the figure, the correlation decay of the V component error is slower than that of the U component. The horizontal decorrelation distance is 50 km, and the vertical decorrelation distance is 1.75 km. In the figure, the yellow solid line and the green dashed line are the negative exponential fitting curves of the correlation coefficient of the V component zonal and meridional errors, respectively, and the red solid line is the negative exponential fitting curve of the vertical error correlation coefficient of the V component, which conforms to the characteristics of a negative exponential function. Figure 5 The figure shows the variation characteristics of the vertical error correlation coefficient of the V component. As can be seen from the figure, the vertical error correlation of the V component gradually decreases within 1.75 km, which clarifies its vertical decorrelation distance.

[0062] The correlation coefficient of the initial field error between points i and j For example, the specific solution process is as follows: Taking the sounding wind field as the true value, for the U component, from Figure 1 The decorrelation distances in the latitudinal and longitudinal directions obtained are 25 km (L). z L m The vertical distance to the relevant location is 1 km (L) h For the V component, L z For 50km, L m For 50km, L h It is 1.75 km. Therefore... The calculation formula is: , where r z r m r h These represent the latitudinal, longitudinal, and perpendicular distances between grid points i and j, respectively. Similarly, the following can be calculated: and .

[0063] Then, local enhancements are applied to key flight areas.

[0064] Point-level and line-level regions were extracted as key flight regions. A high-resolution three-dimensional wind field enhancement field (referred to as the local enhancement field, 0.001°×0.001°×50m, 5 minutes) was generated using the three-dimensional kriging method. Only high-density observation data from the line-level or point-level regions were used, without introducing background field information. A set of observation points was assumed within the study area. and their corresponding wind field values For any interpolation point wind field value :

[0065] in, The wind field estimate for the interpolation point (target wind field grid point) g. For the first i The wind field observations are represented by n, where n is the total number of wind field grid points in the study area. For the first i The interpolation weights for each observation point, to ensure unbiasedness, satisfy the following conditions: The optimal weights are obtained by minimizing the estimated variance, which ultimately boils down to solving the Kriging equations:

[0066] In the formula, Let be the semivariogram (or simply variogram), representing the _th _ i The and the first j A measure of wind field variability among individual wind field grid points; Indicates the first i The and the first g A measure of wind field variability among individual wind field grid points; These are Lagrange multipliers. In three-dimensional space, the variogram is usually assumed to be isotropic, meaning it depends only on the Euclidean distance between two points. .

[0067] The variogram is the core of the Kriging method, reflecting the structural characteristics of spatial variables. Commonly used theoretical models include the spherical model, the exponential model, and the Gaussian model, each suitable for wind fields with different spatial structural characteristics: the spherical model is suitable for wind fields with large-scale uniform variations, the exponential model is suitable for wind fields with drastic local variations, and the Gaussian model is suitable for extremely smooth and highly continuous wind fields. However, key flight areas (such as flight paths and takeoff and landing fields) may simultaneously contain multiple wind field structural characteristics. From high-density building clusters in the city center to open areas in the west of the city, the spatial structure of wind fields varies significantly: wind fields in urban canyon areas vary drastically, while wind fields in open areas are relatively uniform. If a single fixed model is used, it is difficult to adapt to this complex and variable reality, and the interpolation accuracy will be significantly affected. Therefore, this preferred scheme innovatively proposes an adaptive selection strategy for the variogram model, the steps of which are as follows: Step 1: Calculate the experimental variation function. Let there be N observation points within the current local fusion domain, and the distance between any two points i and j be . The corresponding wind field observation values ​​are as follows: and , distance According to step size Grouping, for the k-th group, we have One point pair, Let be the distance from the midpoint of the group, then the experimental variogram value of the k-th group is... for:

[0068] The second step involves performing weighted least-squares fitting of the experimental variation function using a spherical model, an exponential model, and a Gaussian model, respectively. The optimal parameters are then determined by solving the following optimization problem. :

[0069] Where K is the total number of lag distance groups, Weighting coefficients (usually taken as...) (i.e., the number of point pairs in the k-th group). For theoretical models in The values ​​of C0 and a are given, where C0 is the nugget constant, reflecting micro-variation and measurement error; C is the sill value, reflecting structural variation; and a is the range, representing the maximum distance of spatial correlation.

[0070] The third step is to calculate the sum of squared residuals (RSS) for each model. Indicates the optimal parameters Down, The corresponding variation function value of the model.

[0071]

[0072] The fourth step is to select the model with the smallest fitting residual as the variation function model for the current region.

[0073] This strategy calculates the goodness of fit for spherical, exponential, and Gaussian models based on observational data within the current local fusion domain, and automatically selects the optimal model. If the residuals of multiple models are similar (e.g., relative differences less than 5%), the spherical model with fewer parameters and clear physical meaning is preferred. This allows Kriging interpolation to dynamically adapt to different wind field structures, thereby improving the accuracy and adaptability of wind field enhancement in key areas.

[0074] Finally, spatial transition and integration are performed.

[0075] The global base wind field map corresponding to the surface level and the fine-resolution 3D wind field enhancement field corresponding to the point / line level are fused and stitched together using a spatial transition fusion strategy. This step achieves seamless connection and eliminates abrupt changes in the wind field through a formulaic adaptation and smoothing algorithm. First, the overlapping area (with a width of 2-3 grid points) between the global base field and the local enhancement field is identified, and a distance-weighted average algorithm is used for adaptation. The weight allocation formula is based on the distance quantization between the grid point to be fused and the center of the local enhancement field.

[0076]

[0077] in, To enhance the field weight of the three-dimensional wind field with high resolution, As the weight of the global basic wind field base map, d The distance between the grid points of the grid to be fused and the center of the fine-resolution 3D wind field enhancement field (unit: km). d 0 represents the radius of the enhanced three-dimensional wind field region with high resolution (in this embodiment, point level). d 0 = 0.4km, line level d 0 = 0.75km); d 1 represents the distance between the outer boundary of the overlapping region and the center of the high-resolution 3D wind field enhancement field (in this embodiment, point level). d 1 = 0.8km, line level d 1 = 1.5km). Specifically, the weight of the central region of the local enhanced field is set to 75%. At the edge of the overlapping region, the weight of the enhanced field decreases linearly from 75% to 25%, while the weight of the global base field increases linearly from 25% to 75%, thus achieving a gradual transition between the global base field and the local enhanced field and avoiding sudden changes in the wind field.

[0078] For overlapping areas between surface and point levels, as well as between surface and line levels, fusion is achieved using the following formula:

[0079] in, This represents the coordinates of the wind field grid points. Enhanced field grid points for high-resolution 3D wind field The wind field value, Grid points for the global base wind field map The wind field value, To integrate the wind field grid points The wind field value.

[0080] For the edge region (the transition zone at the edge of the fusion boundary, with a width of 2 grid points), a Gaussian smoothing algorithm is used to eliminate abrupt changes in the wind field at the seam. The specific Gaussian smoothing formula is as follows:

[0081] in, Smoothed wind field grid points The wind field value, =0.8, unit is grid points (within the range of 0.5-1.0 grid points, based on Zhejiang Province pilot data calibration). (The radius of the Gaussian kernel is used, and the smoothing window size is 3×3 grid). Represents the grid points of the merged wind field. The wind field value; i and j are the number of grid points moved in the x and y directions, respectively; , The grid spacings in the x and y directions are respectively. This formula eliminates abrupt changes in the wind field at the seams, ensuring the spatiotemporal continuity of the wind field while preserving subtle wind field details such as local wind shear and gusts.

[0082] Outputs a variable-resolution 3D wind field in NetCDF format, including u, v, w wind field components and a resolution identifier field. Update frequency: surface level every 10 minutes, 0.01°×0.01°×200m; point / line level every 5 minutes, 0.001°×0.001°×50m, with a delay of ≤3 minutes.

[0083] To verify the effectiveness of the aforementioned wind field fusion strategy, this invention uses wind field data from April to September 2024 as an example. The method of this invention is tested using three different verification sources, and a comparative test is conducted. The results are shown in Table 1. The root mean square error (RMSE) of the fused wind field using this method in the U and V components is between 2.6 and 3.8 m / s. Using wind profiler radar as an independent verification source, the RMSEs of the fused wind field U and V components are 3.52 m / s and 3.39 m / s, respectively, which are 0.14 m / s and 1.13 m / s lower than the background field (3.66 m / s and 4.52 m / s), respectively. The V component shows a 25% reduction, and the correlation coefficient increases from 0.78 to 0.87, significantly improving wind field accuracy and providing highly reliable three-dimensional wind field data support for low-altitude flight.

[0084] Table 1

[0085] For a case study of Typhoon Bebinca in 2024 (199 time intervals at 6-minute intervals, totaling 2880 samples), the independence of this method was tested using the Jiaxing wind profiler radar. The results are shown in Table 2. Table 2 shows that the root mean square error of the V component of the fused wind field obtained by the method of this invention is 3.32 m / s, while that of the numerical prediction analysis field is 4.21 m / s. The error of the fused wind field is 21% lower than that of the numerical prediction analysis field.

[0086] Table 2

[0087] The results show that the variable resolution 3D wind field generated by this method, through spatial hierarchy partitioning and adaptive fusion strategy, can better serve the application of low-altitude flight in multiple scenarios.

[0088] Example 2 A variable-resolution 3D wind field fusion system for low-altitude flight, the system comprising: The spatial hierarchy division module is used to obtain the take-off and landing point coordinates, flight path and flight airspace range of low-altitude flight missions; and based on the low-altitude flight missions, divide the flight airspace range into three spatial levels: point level, line level and area level. The data acquisition module acquires multi-source wind field observation data within the flight airspace and performs quality control. The global basic wind field base map generation module is used to generate a global basic wind field base map for the surface level by using the optimal interpolation method, taking the numerical forecast analysis field of the flight airspace as the background field, and fusing quality-controlled multi-source wind field observation data. The fine-resolution 3D wind field enhancement field generation module is used to generate fine-resolution 3D wind field enhancement fields corresponding to the point and line levels using the 3D Kriging method. Based on the adaptive selection strategy of the variation function model, the 3D Kriging method is adapted to different wind field structure characteristics. The spatial transition fusion module is used to perform spatial transition fusion between the global basic wind field base map and the fine resolution three-dimensional wind field enhancement field. The fusion is performed in the overlapping areas of the surface layer and the point layer, as well as the surface layer and the line layer. The edge areas of the surface layer and the point layer, as well as the surface layer and the line layer, are processed by the Gaussian smoothing algorithm to obtain a spatially variable resolution three-dimensional wind field.

[0089] Example 3 An electronic device comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the variable resolution three-dimensional wind field fusion method for low-altitude flight.

[0090] Example 4 A computer-readable storage medium having computer instructions stored thereon for causing a computer to perform the steps of the variable-resolution three-dimensional wind field fusion method for low-altitude flight.

Claims

1. A variable-resolution three-dimensional wind field fusion method for low-altitude flight, characterized in that, The steps are as follows: 1) Obtain the coordinates of the take-off and landing points, flight path, and airspace range of the low-altitude flight mission; based on the low-altitude flight mission, divide the airspace range into three spatial levels: point level, line level, and area level. 2) Acquire multi-source wind field observation data within the flight airspace and perform quality control; 3) For the surface level, the optimal interpolation method is adopted, using the numerical forecast analysis field of the flight airspace as the background field, and the quality-controlled multi-source wind field observation data are fused to generate a global basic wind field base map; 4) For line-level and point-level, a three-dimensional kriging method is used to generate fine-resolution three-dimensional wind field enhancement fields corresponding to the point and line levels. Based on the adaptive selection strategy of the variation function model, the three-dimensional kriging method is adapted to different wind field structure characteristics. 5) Spatially transition and fuse the global basic wind field base map described in step 3) with the fine-resolution three-dimensional wind field enhancement field described in step 4). Use a distance-weighted average algorithm to fuse the overlapping areas of the surface layer and the point layer, as well as the surface layer and the line layer. Use a Gaussian smoothing algorithm to process the edge areas of the surface layer and the point layer, as well as the surface layer and the line layer, to obtain a spatially variable resolution three-dimensional wind field.

2. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, In step 1, the three spatial levels—point level, line level, and surface level—correspond to the point level region, line level region, and surface level region, respectively. The criteria for dividing each level are as follows: The surface-level region corresponds to the entire flight airspace range; Within the aforementioned flight airspace: If the distance between any grid point in the target space and any point target is less than the point level threshold, it is divided into a point level region; the point targets include take-off and landing points, regional points in mountains, canyons, and high-altitude areas; If the distance between any grid point in the target space and any line target is less than the line-level threshold, it is divided into a line-level processing area; the line target is the flight path.

3. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, The multi-source wind field observation data includes one or more of the following: wind field retrieved by lidar, wind field retrieved by weather radar, wind field observation data from ground automatic stations, urban canopy ultrasonic wind field, wind tower gradient wind field, or wind field data reported by aircraft.

4. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, The core formula of the three-dimensional kriging method is: in, Let g be the wind field estimate for the target wind field grid point g. Let be the wind field observation value at the i-th wind field grid point, and n be the total number of wind field grid points. For the first The interpolation weights for each observation point satisfy the following condition: ; The optimal weights are obtained by minimizing the estimated variance, specifically by solving the Kriging equations; the Kriging equations are expressed as: In the formula, A measure of wind field variability between the i-th and j-th wind field grid points; A measure of wind field variability between the i-th and g-th wind field grid points; It is a Lagrange multiplier.

5. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, The adaptive selection strategy for the variation function model is as follows: First, calculate the experimental variation function; assume there are N observation points near the current target wind field grid point, and the distance between any two points i and j is... The corresponding wind field observation values ​​are as follows: and , distance According to step size Grouping, for the k-th group, we have One point pair, Let be the distance between the midpoints of the k-th group; then the experimental variogram can be expressed as: in, The value of the variogram for the k-th group; Then, weighted least squares fitting of the experimental variation function was performed using the state model, exponential model, and Gaussian model; the parameters were determined by solving the following optimization problem. The optimal value: Where K is the total number of lag distance groups, These are the weighting coefficients. Let be the experimental variation function value for the k-th group. For spherical models, exponential models, or Gaussian models in The value; , All are parameters of the variogram function, where C0 is the nugget constant, C is the sill value, and a is the range; Next, the sum of squared residuals (RSS) for each model is calculated: in, Indicates the optimal parameters Down, The corresponding variation function value of the model; Finally, the model that minimizes the sum of squared residuals (RSS) is selected as the variation function model for the current region.

6. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, In the overlapping areas of the surface level and the point level, as well as the surface level and the line level, a distance-weighted average algorithm is used for fusion, specifically as follows: in, To integrate the wind field grid points The wind field value, Enhanced field grid points for high-resolution 3D wind field The wind field value, To enhance the field weight of the three-dimensional wind field with high resolution, Grid points for the global base wind field map The wind field value, As the weight of the global basic wind field base map; Weight and The weighting is based on distance quantization between the grid points to be fused and the center of the fine-resolution 3D wind field enhancement field; the weighting formula is: in, The distance between the grid points of the mesh to be fused and the center of the enhanced 3D wind field of high resolution. To enhance the radius of the three-dimensional wind field with high resolution, The distance between the outer boundary of the overlapping region and the center of the enhanced three-dimensional wind field with fine resolution.

7. The variable resolution three-dimensional wind field fusion method for low-altitude flight according to claim 1, characterized in that, Gaussian smoothing is applied to the edge regions between the surface level and the point level, as well as between the surface level and the line level. Specifically: in, Smoothed wind field grid points The wind field value, The value ranges from 0.5 to 1. Let the radius be the Gaussian kernel. Represents the grid points of the merged wind field. The wind field value; i and j are the number of grid points moved in the x and y directions, respectively; , These represent the grid spacing in the x and y directions, respectively.

8. A variable-resolution three-dimensional wind field fusion system for low-altitude flight, characterized in that, include: The spatial hierarchy division module is used to obtain the take-off and landing point coordinates, flight path and flight airspace range of low-altitude flight missions; and based on the low-altitude flight missions, divide the flight airspace range into three spatial levels: point level, line level and area level. The data acquisition module acquires multi-source wind field observation data within the flight airspace and performs quality control. The global basic wind field base map generation module is used to generate a global basic wind field base map for the surface level by using the optimal interpolation method, taking the numerical forecast analysis field of the flight airspace as the background field, and fusing quality-controlled multi-source wind field observation data. The fine-resolution 3D wind field enhancement field generation module is used to generate fine-resolution 3D wind field enhancement fields corresponding to the point and line levels using the 3D Kriging method. Based on the adaptive selection strategy of the variation function model, the 3D Kriging method is adapted to different wind field structure characteristics. The spatial transition fusion module is used to perform spatial transition fusion between the global basic wind field base map and the fine resolution three-dimensional wind field enhancement field. The fusion is performed in the overlapping areas of the surface layer and the point layer, as well as the surface layer and the line layer. The edge areas of the surface layer and the point layer, as well as the surface layer and the line layer, are processed by the Gaussian smoothing algorithm to obtain a spatially variable resolution three-dimensional wind field.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, The computer instructions are used to cause the computer to perform the steps of the method as described in any one of claims 1-7.

Images