WRF mode low-altitude weather forecasting method fusing unmanned aerial vehicle data

Abstract

The invention relates to a WRF mode low-altitude weather forecast method fusing unmanned aerial vehicle data, and belongs to the technical field of weather forecast. The method comprises the steps of data acquisition; preprocessing the data; constructing a WRF initial field; performing dynamic data assimilation; quality control and fusion; and reconstructing and outputting the grid. According to the WRF mode low-altitude weather forecasting method fusing the unmanned aerial vehicle data, the low-altitude weather forecasting error can be reduced, the updating frequency is improved to the minute level, and the WRF mode low-altitude weather forecasting method is particularly suitable for scenes such as urban microclimate monitoring and mountainous area wind power plant site selection.

Description

Technical Field

[0001] The present invention belongs to the technical field of weather forecasting, and in particular relates to a WRF model low-altitude weather forecasting method integrating unmanned aerial vehicle (UAV) data. Background Art

[0002] Meteorological forecasting is a technical system that uses scientific means to predict weather evolution. Its core lies in integrating multi-source observation data with numerical weather prediction (NWP) models to analyze the interaction between atmospheric dynamics and thermodynamics. The current mainstream technology uses meteorological satellites as the core observation carrier. Although it can obtain macroscopic parameters such as global-scale cloud maps, temperature and humidity profiles, its orbital altitude (about 35,786 kilometers) results in limited spatial resolution (usually ≥50 kilometers), making it difficult to capture the dynamic evolution of low-altitude (<1 km) and small-scale weather phenomena (such as low-level jets and thermal convection). Existing technologies have two technical bottlenecks:

[0003] Data source limitations: Relying on satellite and ground-based observations, it is impossible to obtain microphysical parameters such as turbulent kinetic energy and vertical wind shear at high temporal and spatial resolution (minute-level) in the boundary layer (1-3 km);

[0004] Model assimilation flaws: When solely using the WRF (Weather Research and Forecasting) model to assimilate satellite data, the model is limited by parameterization scheme errors (such as boundary layer turbulence closure scheme deviations). This leads to systematic error accumulation in model forecasts in complex terrain areas and makes it impossible to correct parameterization errors in local non-uniform observation data in real time.

[0005] The above-mentioned technical shortcomings make it difficult for existing technologies to meet the needs of high-precision (error <10%) and real-time (delay <1 hour) weather forecasts in low-altitude and small-scale areas (such as urban microclimate zones and mountainous terrain disturbance zones). There is an urgent need to develop innovative solutions that integrate multi-source observation data from drones and adaptive model corrections. Summary of the Invention

[0006] The purpose of the present invention is to provide a low-altitude meteorological forecast method based on the WRF model that integrates UAV data, so as to solve the technical problems that the existing technology relies on satellite observations, resulting in the loss of small-scale meteorological parameters in the low altitude and the accumulation of assimilation errors.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a WRF model low-altitude weather forecasting method integrating drone data, comprising the following steps:

[0009] Data acquisition steps include:

[0010] Receive first meteorological data of the target area released by the GFS global forecast field and second meteorological data and lidar data of the target area collected in real time by a layered drone cluster, wherein the first meteorological data includes temperature data, humidity data and wind field data, and the second meteorological data includes turbulent kinetic energy data, vertical wind shear data and temperature and humidity data;

[0011] Data preprocessing steps include:

[0012] Eliminate outliers that deviate from the main cluster in the first meteorological data and the second meteorological data by using an isolation forest algorithm;

[0013] Using Kriging interpolation to fill missing data in the first meteorological data excluding outliers and the second meteorological data;

[0014] The GDAL library is used to convert the second meteorological data with missing data into NetCDF format, and the WPS tool chain is used to parse the converted NetCDF second meteorological data into a static field.

[0015] The steps for building the initial WRF field include:

[0016] Generate a digital elevation model (DEM) of the target area based on the WPS tool chain and lidar data;

[0017] The improved ICP algorithm is used to dynamically align the second meteorological data and lidar data converted into NetCDF format with the WRF grid in the WPS tool chain to obtain the WRF initial field.

[0018] Dynamic data assimilation steps include:

[0019] The state variables of the initial WRF field are subjected to spatiotemporal weighted filtering by using a spatiotemporal weighted ensemble Kalman filter to obtain multiple assimilated WRF fields. The assimilation window of the spatiotemporal weighted filtering is 15 minutes.

[0020] The assimilated multiple WRF fields are filtered by sliding window mean filtering;

[0021] Quality control and integration steps, including:

[0022] Clustering the filtered multiple WRF fields based on the DBSCAN algorithm, and eliminating outlier WRF fields from the multiple WRF fields;

[0023] The remaining WRF fields after removing the outlier WRF fields from the multiple WRF fields are averaged and weighted by a dynamic Bayesian model to obtain a weighted fused WRF field;

[0024] Mesh reconstruction and output steps include:

[0025] The weighted fused WRF field is divided into multiple layers of grids of different scales, including 0.1 km resolution nested grids and 1 km resolution nested grids;

[0026] The boundary layer parameters of the WRF field after weighted fusion of multiple layers of grids of different scales are optimized through large eddy simulation to obtain a weather forecast model in the target area in GRIB2 format.

[0027] In one possible design, the drone cluster is equipped with an onboard lidar, a micro-meteorological instrument and an infrared camera, which is used to obtain lidar data and turbulent kinetic energy data of the target area through the onboard lidar, obtain target area and vertical wind shear data through the micro-meteorological instrument, and obtain temperature and humidity data of the target area through the infrared camera.

[0028] In one possible design, the improved ICP algorithm introduces a terrain distortion correction term, and the loss function is:

[0029]

[0030] Where λ>0 is the regularization parameter, which is determined by cross-validation. l represents the terrain gradient threshold, which is calculated as l=k·σ, where k is 0.1 to 0.5, and σ is the elevation standard deviation of the lidar data. is the gradient of the residual function R with respect to the transformation parameter t, argmin represents the input that leads to the minimum result, |||| represents the L2 norm, N represents the number of corresponding points between the two sets of point clouds, and q i represents the i-th point in the target point cloud, R(P i +t) represents the transformed source point, and the source point P i The position after translation t and transformation R.

[0031] In the second aspect, the present invention provides a WRF model low-altitude meteorological forecast system that integrates drone data, including: a GFS global forecast field data interface, a layered drone cluster, and a meteorological forecast device. The GFS global forecast field data interface is used to receive the first meteorological data of the target area released by the GFS global forecast field, and the drone cluster is used to collect the second meteorological data and lidar data of the target area in real time. The first meteorological data includes temperature data, humidity data, and wind field data, and the second meteorological data includes turbulent kinetic energy data, vertical wind shear data, and temperature and humidity data. The meteorological forecast device is used to execute the WRF model low-altitude meteorological forecast method that integrates drone data as described in the first aspect or any possible design of the first aspect.

[0032] Beneficial effects:

[0033] The present invention uses drone layered observation and dynamic grid encryption technology to reduce the error of low-altitude meteorological forecasts and increase the update frequency to minute levels. It is particularly suitable for scenarios such as urban microclimate monitoring and wind farm site selection in mountainous areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Flowchart of the WRF model low-altitude weather forecast method that integrates drone data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0036] It should be understood that although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element can be referred to as a second element, and similarly, a second element can be referred to as a first element without departing from the scope of the exemplary embodiments of the present invention.

[0037] Example

[0038] like Figure 1 As shown, the WRF model low-altitude meteorological forecasting method for integrating drone data provided in the first aspect of this embodiment can be executed by, but is not limited to, a computer device with certain computing resources, such as a cloud server, an edge computer equipped with a GPU, a personal computer (PC, a multi-purpose computer with a size, price and performance suitable for personal use; desktops, laptops, small laptops, tablets and ultrabooks are all personal computers), a smart phone, a personal digital assistant (PDA) or a wearable device. Figure 1 As shown, the WRF model low-altitude weather forecast method integrating UAV data may include, but is not limited to, the following steps S101 to S106.

[0039] Step S101: Data acquisition.

[0040] In an embodiment of the present application, a drone cluster can be arranged in layers, and the drone cluster is equipped with equipment such as on-board lidar, micro-meteorological instruments and infrared cameras. When forecasting the meteorological data of the target area, it can receive the first meteorological data of the target area released by the GFS global forecast field and the second meteorological data and lidar data of the target area collected in real time by the layered drone cluster.

[0041] The first meteorological data includes temperature data, humidity data, wind field data, and pressure field data. The size of the target area can be set according to actual conditions. For example, the target area can be a mountainous area or an urban area of 50 km. The temporal resolution of the first meteorological data can be 1 hour, and the horizontal resolution can be 10 km.

[0042] The second meteorological data includes turbulent kinetic energy data, vertical wind shear data, and temperature and humidity data. In the embodiment of the present application, the drone cluster is equipped with equipment such as an onboard laser radar, a micro-meteorological instrument, and an infrared camera. The drone cluster is used to obtain laser radar data and turbulent kinetic energy data of the target area through the onboard laser radar, obtain vertical wind shear data of the target area through the micro-meteorological instrument, and obtain temperature and humidity data of the target area through the infrared camera.

[0043] The drones in a drone swarm are arranged in layers. For example, a drone swarm can consist of 5 drones, arranged in layers at altitudes of 50m, 100m, 150m, 200m, and 250m, with adjacent layers vertically spaced 50m apart. The data return frequency of the drones can be set to be less than or equal to 15 seconds.

[0044] The onboard lidar can be used to obtain turbulent kinetic energy data (vertical contour lines of turbulent kinetic energy) of the target area, with a sampling frequency of 10 Hz, a horizontal resolution of 0.5 m, and a vertical resolution of 0.1 m.

[0045] The micro-meteorological instrument can be used to obtain temperature and humidity data at an altitude of 0-100m in the target area with a time resolution of 1 second.

[0046] The infrared camera can be used to obtain temperature and humidity data near the ground (e.g., 0-300m) in the target area, with a frame rate of 30fps.

[0047] The acquired first meteorological data, second meteorological data, and lidar data can be synchronized with multiple sensors through the GNSS (Global Navigation Satellite System) timing module, with a spatial position error of ≤0.1m.

[0048] Step S102: Data preprocessing.

[0049] In the embodiment of the present application, the data preprocessing step may include, but is not limited to, the following steps S1021-S1023.

[0050] Step S1021: Eliminate abnormal values that deviate from the main cluster in the first meteorological data and the second meteorological data by using the Isolation Forest algorithm.

[0051] Among them, the outlier threshold can be set to deviate from the main cluster by more than 3 times the standard deviation.

[0052] Step S1022: Use Kriging interpolation to fill missing data in the first meteorological data and the second meteorological data after removing outliers.

[0053] Using the Kriging interpolation method, we dynamically assign weights based on the variogram of neighboring points in time and space. For example, if wind speed data at a certain time is missing at a height of 150 meters, we can select neighboring points with a temporal correlation coefficient greater than 0.8 (distance less than 200 meters, time difference less than 5 minutes) for weighted interpolation.

[0054] Step S1023. Convert the second meteorological data with missing data filled in into NetCDF format using the GDAL library, and parse the second meteorological data converted into NetCDF format into a static field using the WPS tool chain.

[0055] In an embodiment of the present application, the JSON format data collected by the drone cluster can be converted into NetCDF format through the GDAL library, and the metadata can be parsed through the ungrib.exe component of the WPS tool chain to generate static field data that conforms to the WRF input specification.

[0056] Step S103: WRF initial field construction.

[0057] In the embodiment of the present application, the WRF initial field construction step may include, but is not limited to, the following steps S1031-S1032.

[0058] Step S1031: Generate a digital elevation model (DEM) of the target area based on the WPS tool chain and lidar data.

[0059] In this embodiment, the WPS toolchain's geogrid.exe component can be used to integrate lidar data collected by a drone swarm to generate a digital elevation model (DEM) with a resolution of ≤10m. For example, in mountainous areas, the DEM resolution can reach 5m, significantly improving the accuracy of terrain simulation of airflow.

[0060] Step S1032. Dynamically align the second meteorological data converted into NetCDF format and the lidar data with the WRF grid in the WPS tool chain using the improved ICP algorithm to obtain the WRF initial field.

[0061] The improved ICP algorithm introduces a terrain distortion correction term, and the loss function is:

[0062]

[0063] Where λ>0 is the regularization parameter, which is determined by cross-validation. l represents the terrain gradient threshold, which is calculated as l=k·σ, where k is 0.1 to 0.5, and σ is the elevation standard deviation of the lidar data. is the gradient of the residual function R with respect to the transformation parameter t, argmin represents the input that leads to the minimum result, |||| represents the L2 norm, N represents the number of corresponding points between the two sets of point clouds, and q i represents the i-th point in the target point cloud, R(P i +t) represents the transformed source point, that is, the source point P i The position after translation t and transformation R.

[0064] Step S104: Dynamic data assimilation.

[0065] In the embodiment of the present application, the dynamic data assimilation step may include but is not limited to the following steps S1041-S1042.

[0066] Step S1041: Use the spatiotemporal weighted ensemble Kalman filter to perform spatiotemporal weighted filtering on the state variables of the WRF initial field to obtain multiple assimilated WRF fields.

[0067] In the embodiment of the present application, a spatiotemporal weighted ensemble Kalman filter (ST-EnKF) can be used, the assimilation window is set to 15 minutes, and the observation error covariance matrix is dynamically calibrated by the UAV data variance. For example, when the standard deviation of the turbulent kinetic energy detected by the UAV cluster is 0.5m 2 / S 2 When , the observation error covariance can be adjusted to 0.5*(0.1+|V|), where V is the wind speed.

[0068] Step S1042: Filter the assimilated multiple WRF fields through sliding window mean filtering.

[0069] The sliding window mean filter (the window size can be 5×5×5 grid) is used to suppress the high-frequency noise in the assimilation process and retain the turbulence-scale fluctuations.

[0070] Step S105: Quality control and fusion.

[0071] In the embodiment of the present application, the quality control and fusion steps may include but are not limited to the following steps S1051-S1052.

[0072] Step S1051: Clustering the filtered multiple WRF fields based on the DBSCAN algorithm, and removing outlier WRF fields from the multiple WRF fields.

[0073] When clustering multiple filtered WRF fields using the DBSCAN algorithm, the neighborhood radius ε can be set to 0.5 km and the minimum number of samples MinPts can be set to 10 to remove outliers from the cluster. For example, in a given assimilation result, three groups of outlier WRF fields (with significantly lower density than neighboring fields) due to sensor failures were identified and removed from the cluster.

[0074] Step S1052: The remaining WRF fields after removing the outlier WRF fields from the multiple WRF fields are averaged and weightedly fused using a dynamic Bayesian model to obtain a weighted fused WRF field.

[0075] When the remaining WRF fields after removing the outlier WRF fields from the multiple WRF fields are averaged and fused by the dynamic Bayesian model, weights can be dynamically assigned based on the residual covariance matrix. The weight formula can be as follows:

[0076]

[0077] Among them, x i is the residual vector of the i-th group of fusion fields, C is the residual covariance matrix, and n represents the total number of the remaining WRF fields.

[0078] Step S106: Mesh reconstruction and output.

[0079] In the embodiment of the present application, the mesh reconstruction and output steps may include but are not limited to the following steps S1061-S1062.

[0080] Step S1061: Divide the weighted fused WRF field into multiple layers of grids of different scales, wherein the multiple layers of grids of different scales include 0.1 km resolution nested grids and 1 km resolution nested grids.

[0081] In this embodiment of the present application, the weighted fused WRF field can be divided into two grids of different scales: low-altitude and high-altitude. The low-altitude domain can use a 0.1km resolution nested grid with 100 vertical layers (interlayer spacing ≤ 10m), and the high-altitude domain can use a 1km resolution nested grid with 20 vertical layers.

[0082] Step S1062: Optimize the boundary layer parameters of the WRF field after weighted fusion of multiple layers of grids of different scales through large eddy simulation to obtain a weather forecast model of the target area in GRIB2 format.

[0083] The WRF model low-altitude meteorological forecast method that integrates drone data provided by the present invention can reduce the error of low-altitude meteorological forecasts through drone layered observation and dynamic grid encryption technology, and increase the update frequency to minute level. It is particularly suitable for scenarios such as urban microclimate monitoring and wind farm site selection in mountainous areas, and can achieve high-precision real-time forecasts of low-altitude small-scale regional meteorological data, providing technical support for new energy development, urban emergency management and aviation safety, and facilitating practical application and promotion.

[0084] The second aspect of an embodiment of the present application provides a WRF model low-altitude meteorological forecast system that integrates drone data, including: a GFS global forecast field data interface, a layered drone cluster, and a meteorological forecast device. The GFS global forecast field data interface is used to receive the first meteorological data of the target area released by the GFS global forecast field, and the drone cluster is used to collect the second meteorological data and lidar data of the target area in real time. The first meteorological data includes temperature data, humidity data, and wind field data, and the second meteorological data includes turbulent kinetic energy data, vertical wind shear data, and temperature and humidity data. The meteorological forecast device is used to execute the WRF model low-altitude meteorological forecast method that integrates drone data as described in the first aspect or any possible design of the first aspect.

[0085] The working process, working details and technical effects of the WRF model low-altitude meteorological forecast system integrating drone data provided in the second aspect of this embodiment can be found in the first aspect of the embodiment and will not be repeated here.

[0086] It should be understood that certain details are provided in the following description to facilitate a thorough understanding of the example embodiments. However, one of ordinary skill in the art will appreciate that the example embodiments can be practiced without these specific details. For example, a system may be shown in block diagrams to avoid obscuring the example with unnecessary detail. In other instances, well-known processes, structures, and techniques may be shown without unnecessary detail to avoid obscuring the example embodiments.

[0087] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A WRF model low-altitude weather forecast method integrating drone data, characterized in that: The steps include: Data acquisition steps include: Receive first meteorological data of the target area released by the GFS global forecast field and second meteorological data and lidar data of the target area collected in real time by a layered drone cluster, wherein the first meteorological data includes temperature data, humidity data and wind field data, and the second meteorological data includes turbulent kinetic energy data, vertical wind shear data and temperature and humidity data; Data preprocessing steps include: Eliminate outliers that deviate from the main cluster in the first meteorological data and the second meteorological data by using an isolation forest algorithm; Using Kriging interpolation to fill missing data in the first meteorological data excluding outliers and the second meteorological data; The GDAL library is used to convert the second meteorological data with missing data into NetCDF format, and the WPS tool chain is used to parse the converted NetCDF second meteorological data into a static field. The steps for building the initial WRF field include: Generate a digital elevation model (DEM) of the target area based on the WPS tool chain and lidar data; The improved ICP algorithm is used to dynamically align the second meteorological data and lidar data converted into NetCDF format with the WRF grid in the WPS tool chain to obtain the WRF initial field. Dynamic data assimilation steps include: The state variables of the initial WRF field are subjected to spatiotemporal weighted filtering by using a spatiotemporal weighted ensemble Kalman filter to obtain multiple assimilated WRF fields. The assimilation window of the spatiotemporal weighted filtering is 15 minutes. The assimilated multiple WRF fields are filtered by sliding window mean filtering; Quality control and integration steps, including: Clustering the filtered multiple WRF fields based on the DBSCAN algorithm, and eliminating outlier WRF fields from the multiple WRF fields; The remaining WRF fields after removing the outlier WRF fields from the multiple WRF fields are averaged and weighted by a dynamic Bayesian model to obtain a weighted fused WRF field; Mesh reconstruction and output steps include: The weighted fused WRF field is divided into multiple layers of grids of different scales, including 0.1 km resolution nested grids and 1 km resolution nested grids; The boundary layer parameters of the WRF field after weighted fusion of multiple layers of grids of different scales are optimized through large eddy simulation to obtain a weather forecast model in the target area in GRIB2 format.

2. The WRF model low-altitude meteorological forecast method integrating UAV data according to claim 1 is characterized in that: The drone cluster is equipped with a laser radar, a micro-meteorological instrument and an infrared camera, which is used to obtain laser radar data and turbulent kinetic energy data of the target area through the laser radar, obtain vertical wind shear data of the target area through the micro-meteorological instrument, and obtain temperature and humidity data of the target area through the infrared camera.

3. The WRF model low-altitude meteorological forecast method integrating UAV data according to claim 1 is characterized in that: The improved ICP algorithm introduces a terrain distortion correction term, and the loss function is: Where λ>0 is the regularization parameter, which is determined by cross-validation. l represents the terrain gradient threshold, which is calculated as l=k·σ, where k is 0.1 to 0.5, and σ is the elevation standard deviation of the lidar data. is the gradient of the residual function R with respect to the transformation parameter t, argmin represents the input that leads to the minimum result, || || represents the L2 norm, N represents the number of corresponding points between the two sets of point clouds, and q i represents the i-th point in the target point cloud, R(P i +t) represents the transformed source point, and the source point P i The position after translation t and transformation R.

4. A WRF model low-altitude weather forecast system integrating drone data, characterized by: include: A GFS global forecast field data interface, a layered drone cluster, and a meteorological forecast device, wherein the GFS global forecast field data interface is used to receive the first meteorological data of the target area released by the GFS global forecast field, and the drone cluster is used to collect the second meteorological data and lidar data of the target area in real time, wherein the first meteorological data includes temperature data, humidity data, and wind field data, and the second meteorological data includes turbulent kinetic energy data, vertical wind shear data, and temperature and humidity data, and the meteorological forecast device is used to execute the WRF model low-altitude meteorological forecast method for integrating drone data as described in any one of claims 1 to 3.

Images