A weather radar three-dimensional wind field intelligent fusion inversion system and method

Abstract

The invention relates to the technical field of wind field inversion, and discloses a weather radar three-dimensional wind field intelligent fusion inversion system and method, comprising: a data processing module, the data processing module comprising a decoding unit and a data quality control unit; the decoding unit is used to decode the wind field product base data; the data quality control unit is used to interpolate the basic reflectivity and radial velocity in the wind field product base data to the same grid field after quality control; the weather radar three-dimensional wind field intelligent fusion inversion system and method, by fusing numerical forecast and optical flow wind field as the initial wind field, the cost function consists of radial velocity observation constraint, mass conservation constraint, background constraint, etc., and the localization weight coefficient is determined by effect test through a severe convective weather case set, the wind field inversion is performed in a multi-radar common observation area, and a three-dimensional variational inversion algorithm for the wind field is realized based on GPU parallel computing technology, the inversion time is within 2 minutes, and the demand for business operation is met.

Description

Technical Field

[0001] The present invention relates to the technical field of wind field inversion, and in particular to a weather radar three-dimensional wind field intelligent fusion inversion system and method. Background Art

[0002] At present, the research on three-dimensional variational wind field inversion technology by experts and scholars at home and abroad is usually based on dual (multi) radar data, and most of the research focuses on case studies of severe convective weather. Since the three-dimensional wind field inversion process based on variational technology is computationally intensive, its algorithm often fails to meet the requirements of commercial operation (inversion time is more than 60 minutes). Summary of the invention

[0003] In order to solve the above problem that the current research on three-dimensional variational wind field inversion technology by domestic and foreign experts and scholars is usually based on dual (multi) radar data, and its algorithm often fails to meet the needs of commercial operation, the present invention is implemented through the following technical solutions: A weather radar three-dimensional wind field intelligent fusion inversion system, comprising:

[0004] A data processing module, wherein the data processing module comprises a decoding unit and a data quality control unit;

[0005] The decoding unit is used to decode the wind farm product base data;

[0006] The data quality control unit is used to interpolate the basic reflectivity and radial velocity in the wind field product base data to the same grid field after quality control;

[0007] A wind field inversion module, wherein the wind field unit module includes an initial wind field calculation unit, a three-dimensional wind field solution unit and a wind field product generation unit;

[0008] The initial wind field calculation unit calculates the initial wind field using an optical flow method;

[0009] The three-dimensional wind field solving unit uses GPU parallel computing technology to iteratively solve the three-dimensional wind field;

[0010] The wind farm product generation unit is used to generate wind farm products;

[0011] The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell throughout its life cycle.

[0012] The S-band radar observation data and X-band phased array radar observation data in the "Severe Convective Weather Case Collection" were collected and screened. Based on the 100-meter scale three-dimensional wind field fused by large and small radars, the bottom height, top height, diameter, maximum tangential shear and other characteristics of the mesocyclone products were identified based on single radar observations. The vorticity, lifting helicity, storm relative helicity and other related flow field parameters of the storm cell were calculated, and the characteristic quantities of the mesocyclone were tracked and statistically analyzed to obtain the characteristics of the entire life cycle of the supercell.

[0013] The calculation formula of lift helicity (UH) is:

[0014]

[0015] Where w is the vertical velocity, is the vertical component of relative vorticity, Z0 and Zt are heights, usually Z0 is taken as 2km and Zt is taken as 5km;

[0016] The calculation formula for the relative helicity of a storm is:

[0017]

[0018] Among them, V is the ambient wind speed, C is the storm moving speed, Ω xy is the horizontal vorticity vector, Z is the storm inflow thickness, and Z is usually 1km or 3km.

[0019] Furthermore, the basic data of the wind farm product is basic data of several X-band phased array radars and several S-band weather radars.

[0020] Further, the data processing module is a multi-process data processing program, including process 1, process 2, ..., process N, and the data processing module also includes a file monitoring unit and a temporary data file generating unit;

[0021] The file monitoring unit is used to perform data file monitoring on the wind farm product base data;

[0022] The temporary data file production unit is used to generate a temporary data file from the data product processed by the data quality control unit.

[0023] Furthermore, the data quality control unit uses a velocity defuzzification algorithm and an isolated small echo identification and filtering algorithm to perform data quality control on the basic data of the wind farm product;

[0024] The velocity deambiguation algorithm is Region-based Recursive Doppler Dealiasing (R2D2) and Four-dimensional Dealiasing (4DD) based on the continuity principle of radial velocity in space and time. The above velocity deambiguation algorithm is highly efficient. The selection of velocity deambiguation algorithm: first, collect and screen the severe convective weather case set observed by S-band weather radar within the wind field inversion range in the past two years, including mesocyclone weather; then, statistically evaluate the application effects of multiple commonly used deambiguation algorithms on the "severe convective weather case set" of X-band phased array radar, so as to select the radial velocity deambiguation algorithm suitable for multi-band weather radar;

[0025] The application of the isolated small echo identification and filtering algorithm is as follows: first, the basic reflectivity and radial data of the wind field product base data are interpolated to the Cartesian coordinates, and then the "centrosymmetric matrix eigenvalue and eigenvector" algorithm is applied to identify the reflectivity echo of each radar volume scan, and the radial wind at the corresponding position is marked and filtered.

[0026] Furthermore, the initial wind field calculation unit calculates the initial wind field specifically as follows:

[0027] The inversion range is divided into several regions, and the basic reflectivity of the radar for multiple continuous time series is selected in each region. The Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region.

[0028] The Lucas-Kanande pyramid optical flow method introduces pyramid layering, applies affine transformation to improve the Lucas-Kanade algorithm, and the calculation steps include: firstly calculating the optical flow of the topmost layer image of the pyramid, calculating the initial value of the optical flow of the next upper layer according to the optical flow result of the topmost layer, and then further estimating the precise value of the optical flow, and finally estimating the initial value of the optical flow of the next layer by using the calculated optical flow result of the next upper layer, and continuing to substitute the precise value into the next layer for calculation until the calculation reaches the bottom layer of the pyramid.

[0029] Further, the three-dimensional wind field solving unit is based on the weighted localization calculation method of the cost function coefficient, uses the three-dimensional variational method, takes the radial velocity of the networked radar as the basis, fuses the data of numerical forecast, wind profile radar and ground observation, and obtains the inverted wind field after multiple constraint iterations. The wind field inversion scheme is based on the wind field inversion technology principle of multi-source data fusion. The cost function J includes radial velocity observation constraint Jo, mass conservation constraint Jm, bottom boundary layer constraint Jp, smoothness constraint Js, background field constraint Jb and point constraint JPoint. The background field constraint uses the wind direction and speed data of the meteorological automatic station as the background field, and the point constraint uses the wind profile radar data. The optimal wind field solution is at the (global) minimum value of the cost function. The cost function J is:

[0030] J=Jo+Jm+Jp+Js+Jb+JPoint;

[0031] Among them, the radial velocity observation constraint Jo is defined as:

[0032]

[0033] The mass conservation constraint Jm is defined as:

[0034]

[0035] The bottom boundary layer constraint Jp is defined as:

[0036]

[0037] The smoothness constraint Js is defined as:

[0038]

[0039] Among them, u, v, w are three-dimensional wind fields, Rad is radial, and Cart is a classification regression tree. To observe the speed, is the radial velocity, λ o1 , o2 ,λm,λ s1 , s2 , s3 , s4 are all constants, ρ is the air density, are the boundary rotation angular velocities of the air mass relative to the x-axis, y-axis, and z-axis in the three-dimensional wind field, respectively; T is the temperature, ω is the coefficient weight, and ^p is the diagonal matrix of the coefficient weight.

[0040] Furthermore, the three-dimensional wind field solving unit uses GPU parallel computing technology based on a deep learning framework to iteratively calculate the three-dimensional inversion process to shorten the inversion time and improve the inversion speed.

[0041] Furthermore, the wind field product generated by the wind field product generating unit is a three-dimensional wind field graphic product, and the three-dimensional wind field graphic product includes a three-dimensional wind field, vorticity, divergence, strong wind, rising helicity and storm relative helicity.

[0042] Furthermore, it also includes an offshore typhoon positioning module, which calculates the relative vorticity of 10m, 850hPa, 700hPa, and 500hPa and the center of the wind field to locate the typhoon center based on the three-dimensional wind field and GVT technical method.

[0043] A weather radar three-dimensional wind field intelligent fusion inversion method, including the following specific methods:

[0044] Decoding the wind farm product base data using the decoding unit;

[0045] The velocity defuzzification algorithm and isolated small echo identification and filtering algorithm are used to perform data quality control on the basic data of wind farm products;

[0046] The inversion range is divided into several regions, and the basic reflectivity of the radar for multiple continuous time series is selected in each region. The Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region.

[0047] Based on the weighted localization calculation method of the cost function coefficient, the three-dimensional variational method is used, based on the radial velocity of the networked radar, and the inverted wind field is obtained by fusing the data of numerical forecast, wind profile radar and ground observation, and after multiple constraint iterations.

[0048] GPU parallel computing technology based on deep learning framework is used to iterate the three-dimensional inversion process to generate wind field products;

[0049] The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell throughout its life cycle.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The weather radar three-dimensional wind field intelligent fusion inversion system and method, by fusing numerical forecasts and optical flow wind fields as the initial wind field, the cost function is composed of radial velocity observation constraints, mass conservation constraints, background constraints, smoothness constraints, point constraints and ground boundary layer constraints, and the localization weight coefficient is determined by testing the effect through a set of severe convective weather cases, and the wind field inversion is performed in the common observation area of ​​multiple radars, and the three-dimensional variational inversion algorithm of the wind field is realized by using GPU-based parallel computing technology. The target inversion time is within 2 minutes, meeting the requirements of commercial operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The process of the weather radar three-dimensional wind field intelligent fusion inversion system of the present invention is as follows Figure 1 ;

[0053] Figure 2 The process of the weather radar three-dimensional wind field intelligent fusion inversion system of the present invention is as follows Figure 2 . DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The embodiments of the weather radar three-dimensional wind field intelligent fusion inversion system and method are as follows:

[0056] Among them, the inversion range of wind field products is centered on 22.45°N, 112.75°E, with east-west and north-south diameters of 1000km each, a vertical range of 0-10km, and a temporal and spatial resolution of: 1km*1km*500m, 6min.

[0057] See also Figure 1-Figure 2 , a weather radar three-dimensional wind field intelligent fusion inversion system, including:

[0058] Data processing module, the data processing module is a multi-process data processing program, including process 1, process 2, ..., process N, the data processing module includes a file monitoring unit, a decoding unit, a data quality control unit and a temporary data file generation unit;

[0059] The file monitoring unit is used to monitor the data files of the wind farm product base data;

[0060] The decoding unit is used to decode the base data of wind field products. The base data of wind field products are the base data of several X-band phased array radars and several S-band weather radars. Several S-band weather radars take data from 20 S-band weather radars in Guangdong and surrounding provinces and apply them to S-band weather radar three-dimensional wind field inversion products and wind field secondary products. Guangdong and surrounding provinces include Guangzhou, Shenzhen, Shaoguan, Qingyuan, Zhanjiang, Shantou, Haikou and Longyan;

[0061] The data quality control unit is used to interpolate the basic reflectivity and radial velocity in the wind farm product base data to the same grid field after quality control; the data quality control unit uses velocity defuzzification algorithm and isolated small echo identification and filtering algorithm to perform data quality control on the wind farm product base data.

[0062] The velocity deambiguation algorithms are Region-based Recursive Doppler Dealiasing (R2D2) and Four-dimensional Dealiasing (4DD) based on the continuity principle of radial velocity in space and time. The above velocity deambiguation algorithms are efficient. The selection of velocity deambiguation algorithms is as follows: First, the relevant data of mesocyclones and TVS weather cases in Foshan area observed by Guangzhou S-band weather radar within the wind field inversion range in the past two years (severe convective weather case set) are collected and screened; then, the application effects of the four commonly used deambiguation algorithms on the "severe convective weather case set" of X-band phased array radar are statistically evaluated. The four deambiguation algorithms include: four-dimensional dealiasing, region based, multipass and unwrap, so as to select the radial velocity deambiguation algorithm suitable for multi-band weather radar.

[0063] The application of the isolated small echo identification and filtering algorithm is as follows: first, the basic reflectivity and radial data of the wind field product base data are interpolated to the Cartesian coordinates, and then the "centrosymmetric matrix eigenvalue and eigenvector" algorithm is applied to identify the reflectivity echo of each radar volume scan, and the radial wind at the corresponding position is marked and filtered.

[0064] Wind field inversion module, the wind field unit module includes an initial wind field calculation unit, a three-dimensional wind field solution unit and a wind field product generation unit;

[0065] The temporary data file production unit is used to generate temporary data files from the data products processed by the data quality control unit.

[0066] The initial wind field calculation unit calculates the initial wind field using the optical flow method; the initial wind field calculation unit calculates the initial wind field specifically as follows:

[0067] The inversion range of wind field products is divided into six regions. The basic reflectivity of three continuous time series of radar is selected in each region, and the Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region.

[0068] The Lucas-Kanande pyramid optical flow method introduces pyramid layering and applies affine transformation to improve the Lucas-Kanade algorithm. The calculation steps include: first calculating the optical flow of the top-level image of the pyramid, calculating the initial value of the optical flow of the next upper layer based on the optical flow result of the top layer, and then further estimating the precise value of the optical flow. Finally, the calculated optical flow result of the next upper layer is used to estimate the initial value of the optical flow of the next layer. After calculating its precise value, it is substituted into the next layer for calculation until the calculation reaches the bottom layer of the pyramid.

[0069] The three-dimensional wind field solving unit uses GPU parallel computing technology to iteratively solve the three-dimensional wind field; the three-dimensional wind field solving unit is based on the weighted localized calculation method of the cost function coefficient, uses the three-dimensional variational method, and is based on the radial velocity of the networked radar. By fusing the data of numerical forecast, wind profile radar and ground observation, and after multiple constraint iterations, the inverted wind field is obtained. The wind field inversion scheme is based on the wind field inversion technology principle of multi-source data fusion. This embodiment is based on the Guangdong wind field inversion scheme based on multi-source data fusion: 17 SA radars (11 in Guangdong, 3 in Guangxi, and 3 in Hainan) near the Guangdong region are used to carry out Guangdong three-dimensional wind field inversion. The cost function J includes radial velocity observation constraints Jo, mass conservation constraints Jm, bottom boundary layer constraints Jp, smoothness constraints Js, background field constraints Jb and point constraints JPoint. The optimal wind field solution is at the (global) minimum value of the cost function, where the initial wind field uses numerical forecasts, the background field constraints use the wind direction and wind speed data of the meteorological automatic station as the background field, and the point constraints use wind profile radar data. The cost function J is:

[0070] J=Jo+Jm+Jp+Js+Jb+JPoint;

[0071] This embodiment is based on a 100-meter-scale wind field inversion scheme based on the fusion of large and small radars: the area near the Pearl River Delta Hub (New Guangzhou) Airport is selected as the test area for the fusion inversion of the three-dimensional wind field of the large and small radars, and the three-dimensional wind field inverted by the SA radar is used as the initial wind field of the test area. Then, a high-precision inversion test of the storm is carried out using the X-band phased array radar near the airport, thereby obtaining the detailed three-dimensional wind field structure and products in the key areas, which helps to gain a deeper understanding of the dynamic structure of convective storms.

[0072] Among them, the radial velocity observation constraint Jo is defined as:

[0073]

[0074] The mass conservation constraint Jm is defined as:

[0075]

[0076] The bottom boundary layer constraint Jp is defined as:

[0077]

[0078] The smoothness constraint Js is defined as:

[0079]

[0080] Among them, u, v, w are three-dimensional wind fields, Rad is radial, and Cart is a classification regression tree. To observe the speed, is the radial velocity, λ o1 ,o2 ,λm,λ s1 , s2 , s3 , s4 are all constants, ρ is the air density, are the boundary rotation angular velocities of the air mass relative to the x-axis, y-axis, and z-axis in the three-dimensional wind field, T is the temperature, ω is the coefficient weight, ^ p is the diagonal matrix of coefficient weights.

[0081] This embodiment selects 1,000 radar data points from 6 weather processes in 2021 (Guangzhou, Zhaoqing, and Shenzhen) for wind field inversion. 30 coefficient weight combinations are applied for calculation at each time point, and the mean square errors of 30,000 three-dimensional wind field radial projections in the Yangjiang and Heyuan radars and the radial velocities of the Yangjiang and Heyuan radars are statistically analyzed to determine the coefficient weights.

[0082] The 3D wind field solution unit uses GPU parallel computing technology based on a deep learning framework to iterate the 3D inversion process, ensuring that the wind field inversion time is less than 1 minute and the output product is within 5 minutes, which is used to shorten the inversion time and increase the inversion speed. GPU parallel computing supports data from multiple types of radars, including phased array radars from manufacturers such as SA, Zhuhai-Macao Radar, NanoRay and Rayleigh, and GPU parallel computing technology can meet the needs of business operations.

[0083] The wind field product generation unit is used to generate wind field products. The wind field products generated by the wind field product generation unit are three-dimensional wind field graphic products. The three-dimensional wind field graphic products include three-dimensional wind field, vorticity, divergence, strong winds, rising helicity and storm relative helicity.

[0084] The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell throughout its life cycle.

[0085] This embodiment collects and screens the Guangzhou S-band weather radar observation data and Foshan X-band phased array radar observation data in the "Severe Convective Weather Case Collection", based on the 100-meter scale three-dimensional wind field fused by large and small radars, combined with the bottom height, top height, diameter, maximum tangential shear and other characteristics of the mesocyclone product identified based on a single radar observation, calculates the vorticity, lifting helicity, storm relative helicity and other related flow field parameters of the storm cell, and tracks and statistically analyzes the characteristic quantities of the mesocyclone to obtain the full life cycle characteristics of the supercell.

[0086] The calculation formula of lift helicity (UH) is:

[0087]

[0088] Where w is the vertical velocity, is the vertical component of relative vorticity, Z0 and Zt are heights, usually Z0 is taken as 2km and Zt is taken as 5km;

[0089] The calculation formula for the relative helicity of a storm is:

[0090]

[0091] Among them, V is the ambient wind speed, C is the storm moving speed, Ω xy is the horizontal vorticity vector, Z is the storm inflow thickness, and Z is usually 1km or 3km.

[0092] The offshore typhoon positioning module calculates the relative vorticity of 10m, 850hPa, 700hPa, and 500hPa and the center of the wind field to locate the typhoon center based on the three-dimensional wind field and GVT technical method.

[0093] GVT method: Apply numerical forecasts to determine the typhoon center by searching for the location of the maximum or minimum value of the variable, giving priority to the 10m, 850hPa, 700hPa, 500hPa relative vorticity, the low center of sea level pressure, the 850hPa, 700hPa, 500hPa geopotential height field center, followed by the 10m, 850hPa, 700hPa, 500hPa wind field center.

[0094] This embodiment determines the typhoon center based on the three-dimensional wind field in Guangdong region and the GVT technical method to calculate the relative vorticity and wind field center at 10m, 850hPa, 700hPa, and 500hPa.

[0095] A weather radar three-dimensional wind field intelligent fusion inversion method, including the following specific methods:

[0096] Decoding the wind farm product base data using a decoding unit;

[0097] The velocity defuzzification algorithm and isolated small echo identification and filtering algorithm are used to perform data quality control on the basic data of wind farm products;

[0098] The inversion range is divided into several regions, and the basic reflectivity of the radar for multiple continuous time series is selected in each region. The Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region.

[0099] Based on the weighted localization calculation method of the cost function coefficient, the three-dimensional variational method is used, based on the radial velocity of the networked radar, and the inverted wind field is obtained by fusing the data of numerical forecast, wind profile radar and ground observation, and after multiple constraint iterations.

[0100] GPU parallel computing technology based on deep learning framework is used to iterate the three-dimensional inversion process to generate wind field products;

[0101] The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell's entire life cycle.

[0102] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A weather radar three-dimensional wind field intelligent fusion inversion system, characterized in that: include: A data processing module, wherein the data processing module includes a decoding unit and a data quality control unit; The decoding unit is used to decode the wind farm product base data; The data quality control unit is used to interpolate the basic reflectivity and radial velocity in the wind field product base data to the same grid field after quality control; A wind field inversion module, the wind field inversion module comprising an initial wind field calculation unit, a three-dimensional wind field solution unit and a wind field product generation unit; The initial wind field calculation unit calculates the initial wind field using an optical flow method; The three-dimensional wind field solving unit uses GPU parallel computing technology to iteratively solve the three-dimensional wind field; The wind farm product generation unit is used to generate wind farm products; The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell throughout its life cycle.

2. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 1 is characterized by: The wind farm product basic data is the basic data of several X-band phased array radars and several S-band weather radars.

3. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 1 is characterized by: The data processing module is a multi-process data processing program, and the data processing module also includes a file monitoring unit and a temporary data file generation unit; The file monitoring unit is used to perform data file monitoring on the wind farm product base data; The temporary data file production unit is used to generate a temporary data file from the data product processed by the data quality control unit.

4. The weather radar three-dimensional wind field intelligent fusion inversion system according to any one of claims 1 to 3, characterized in that: The data quality control unit uses a velocity defuzzification algorithm and an isolated small echo identification and filtering algorithm to perform data quality control on the basic data of the wind farm product; The velocity deblurring algorithm is Region-based Recursive Doppler Dealiasing and Four-dimensional Dealiasing based on the continuity principle of radial velocity in space and time; The application of the isolated small echo identification and filtering algorithm is as follows: first, the basic reflectivity and radial data of the wind farm product base data are interpolated to the Cartesian coordinates, and then the "centrosymmetric matrix eigenvalue and eigenvector" algorithm is applied to identify the reflectivity echo of each radar volume scan, and the radial wind at the corresponding position is marked and filtered.

5. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 4 is characterized by: The initial wind field calculation unit calculates the initial wind field specifically as follows: The inversion range is divided into several regions, and the basic reflectivity of the radar for multiple continuous time series is selected in each region. The Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region. The Lucas-Kanande optical flow pyramid algorithm introduces pyramid layering, applies affine transformation to improve the Lucas-Kanade algorithm, and the calculation steps include: firstly calculating the optical flow of the topmost layer image of the pyramid, calculating the initial value of the optical flow of the next upper layer according to the optical flow result of the topmost layer, and then further estimating the precise value of the optical flow, and finally estimating the initial value of the optical flow of the next layer by using the calculated optical flow result of the next upper layer, and continuing to substitute the precise value into the next layer for calculation until the calculation reaches the bottom layer of the pyramid.

6. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 5 is characterized by: The three-dimensional wind field solving unit is based on the weighted localization calculation method of the cost function coefficient, uses the three-dimensional variational method, takes the radial velocity of the networked radar as the basis, fuses the data of numerical forecast, wind profile radar and ground observation, and obtains the inverted wind field through multiple constraint iterations. The cost function J includes radial velocity observation constraint Jo, mass conservation constraint Jm, bottom boundary layer constraint Jp, smoothness constraint Js, background field constraint Jb and point constraint JPoint. The background field constraint uses the wind direction and wind speed data of the meteorological automatic station as the background field, and the point constraint uses the wind profile radar data, that is, J=Jo+Jm+Jp+Js+Jb+JPoint.

7. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 6 is characterized by: The three-dimensional wind field solving unit uses GPU parallel computing technology based on a deep learning framework to iterate the three-dimensional inversion process.

8. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 7 is characterized by: The wind field product generated by the wind field product generating unit is a three-dimensional wind field graphic product, and the three-dimensional wind field graphic product includes a three-dimensional wind field, vorticity, divergence, strong wind, rising helicity and storm relative helicity.

9. The weather radar three-dimensional wind field intelligent fusion inversion system according to claim 8 is characterized by: It also includes an offshore typhoon positioning module, which calculates the relative vorticity of 10m, 850hPa, 700hPa, and 500hPa and the center of the wind field to locate the typhoon center based on the three-dimensional wind field and GVT technical method.

10. A weather radar three-dimensional wind field intelligent fusion inversion method, using the weather radar three-dimensional wind field intelligent fusion inversion system according to claim 9, characterized in that: The following specific methods are included: Decoding the wind farm product base data using the decoding unit; The velocity defuzzification algorithm and isolated small echo identification and filtering algorithm are used to perform data quality control on the basic data of wind farm products; The inversion range is divided into several regions, and the basic reflectivity of the radar for multiple continuous time series is selected in each region. The Lucas-Kanande optical flow pyramid algorithm is used to calculate the initial wind field of each region. Based on the weighted localization calculation method of the cost function coefficient, the three-dimensional variational method is used, based on the radial velocity of the networked radar, and the inverted wind field is obtained by fusing the data of numerical forecast, wind profile radar and ground observation, and after multiple constraint iterations. GPU parallel computing technology based on deep learning framework is used to iterate the three-dimensional inversion process to generate wind field products; The supercell flow field feature extraction module is used to identify the characteristics of the bottom height, top height, diameter, and maximum tangential shear of the mesocyclone product, calculate the parameters of the storm cell vorticity, lifting helicity, and storm relative helicity, and track and statistically analyze the mesocyclone characteristic quantities to obtain the characteristics of the supercell throughout its life cycle.

Images