A sky weather forecasting system based on real-time airspace correction of unmanned aerial vehicles

Abstract

The application belongs to the technical field of meteorological monitoring, and provides a weather perception system of Tianhai Zhi Mo based on real-time correction of airspace weather forecast by unmanned aerial vehicle. An air-sea-land intelligent coupling mechanism is established, real-time fusion of meteorological ground observation, ocean circulation field and unmanned aerial vehicle detection data is carried out based on ensemble Kalman filtering algorithm, and an efficient three-dimensional wind field reconstruction engine is constructed. A physical enhancement AI correction strategy is adopted, an LSTM-Transformer hybrid model is used to drive forecast update, and a gradient constraint mechanism is introduced to effectively inhibit the non-physical mutation of the meteorological field. The final value of the system is reflected in the depth of the airspace control synergy, which can automatically generate a risk heat map, plan an obstacle avoidance path, and trigger control instructions in conjunction with the air traffic management system. The comprehensive application of the system breaks through the limitations of traditional methods in time and space resolution, and provides high time and space resolution early warning support for port scheduling, unmanned aerial vehicle logistics and other scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological monitoring and relates to a Tianhai Zhimou meteorological perception system based on real-time correction of airspace weather forecast by unmanned aerial vehicles. Background Art

[0002] With the rapid development of the low-altitude economy, emerging industries such as drone logistics and urban air traffic are placing higher demands on the refinement and real-time nature of airspace meteorological services. Traditional meteorological monitoring relies on satellite remote sensing and sparsely distributed ground-based observation stations, which have insufficient spatial resolution and are unable to capture key meteorological phenomena such as micro-scale turbulence and localized severe convection in complex terrain. Numerical forecast models, due to their high computational complexity and long update cycles of several hours, cannot meet the minute-by-minute risk avoidance needs of low-altitude aircraft. While unmanned aerial vehicle (UAV) mobile platforms can fill observation gaps, existing technologies often focus on collecting a single atmospheric parameter, failing to achieve deep coupling of air, sea, and land data, making it difficult to track the detailed evolution of severe convection.

[0003] Current technical bottlenecks primarily arise from the disconnected processing of ocean circulation data and atmospheric sounding data, which results in distorted simulations of air-sea interaction processes. Furthermore, the high-dimensional, heterogeneous data transmitted by drones lacks an efficient assimilation mechanism, and traditional Kalman filters are inadequate for processing three-dimensional wind fields that update at the second level. Furthermore, AI-powered meteorological models exhibit significant physical distortion when applied to complex terrain, and their warning results have yet to be effectively integrated with airspace control systems. While CN120044638 addresses drone-based meteorological monitoring, it fails to establish a dynamic closed-loop "detection-correction-redetection" mechanism. Purely data-driven models, such as CNN-LSTM, exhibit high false alarm rates in mountainous areas.

[0004] To address these challenges, the present invention innovatively utilizes drones to revise airspace weather forecasts in real time: the system uses a swarm of drones equipped with Doppler lidar and microwave radiometers to achieve high-temporal-resolution scanning of turbulence structures and temperature and humidity gradients in low-altitude areas below 1,000 meters. The core technological breakthrough lies in the establishment of an air-sea-land intelligent coupling mechanism, which uses the ensemble Kalman filter algorithm to integrate meteorological ground-based observations, ocean circulation fields, and drone detection data in real time to construct an efficient three-dimensional wind field reconstruction engine. It also adopts a physically enhanced AI correction strategy, uses the LSTM-Transformer hybrid model to drive forecast updates, and introduces a gradient constraint mechanism to effectively suppress non-physical mutations in the meteorological field. The ultimate value of the system is reflected in the deep coordination of airspace control. It can automatically generate risk heat maps, plan obstacle avoidance paths, and link with the air traffic management system to trigger control instructions. The comprehensive application of this system breaks through the limitations of traditional methods in temporal-spatial resolution, providing high-temporal-spatial resolution early warning support for scenarios such as port scheduling and drone logistics. Summary of the Invention

[0005] The purpose of the embodiment of the present invention is to provide a Tianhai Zhimou meteorological perception system based on real-time correction of airspace weather forecasts by drones, aiming to solve the problems raised in the background technology.

[0006] The technical solution of the present invention:

[0007] A Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecast by drones has the following steps:

[0008] (1) Constructing a dynamic networking unit for drone swarms;

[0009] The UAV swarm dynamic networking unit is equipped with a Doppler laser radar, microwave radiometer and three-dimensional ultrasonic anemometer on the drone. The Doppler laser radar, microwave radiometer and three-dimensional ultrasonic anemometer are used to perform three-dimensional scanning of the low-altitude area and collect three-dimensional airspace micro-meteorological data in real time.

[0010] Choose a product with high load capacity ( 5kg) medium and large rotor drones, ensuring they can carry multiple sensors;

[0011] The Doppler lidar is installed on the belly gimbal of the drone;

[0012] The microwave radiometer is deployed on the top of the drone's fuselage and operates at a frequency of 23.8 GHz. 25MHz (temperature and humidity inversion);

[0013] The three-dimensional ultrasonic anemometer is distributed and installed at the end of the UAV's rotor arm (range 0-60m / s, sampling frequency 10Hz);

[0014] An edge computing module is used to compress three-dimensional airspace micrometeorological data in real time; the edge computing module mainly consists of a processor, memory, and communication interface;

[0015] (2) Build a land-sea-air multi-source data fusion engine to fuse meteorological ground-based observation data, ocean circulation field data, and UAV detection data based on the ensemble Kalman filter algorithm;

[0016] Real-time acquisition of meteorological ground-based observation data from ground stations across the country, including temperature, humidity, and air pressure data, which are then connected to the ensemble Kalman filter algorithm through the API protocol;

[0017] Ocean circulation data are obtained through international open data platforms, including but not limited to sea surface temperature and current data from CMEMS, NASA PODAAC, or NOAA OISST;

[0018] The meteorological ground-based observation data, ocean circulation data and UAV detection data are fused using the ensemble Kalman filter algorithm to output the 3D wind field reconstruction result.

[0019] (3) The LSTM-Transformer hybrid model is used to dynamically correct the 3D wind field reconstruction results, and the accuracy of the LSTM-Transformer hybrid model is optimized by combining the gradient constraint mechanism;

[0020] Build an LSTM-Transformer hybrid model to process time-series meteorological data. The LSTM-Transformer hybrid model includes a long short-term memory network, a self-attention mechanism, a spatiotemporal graph convolutional network, and a temporal convolutional layer. The self-attention mechanism includes an encoder and a decoder. The time-series meteorological data includes ground-based meteorological observation data, ocean circulation data, and three-dimensional wind field reconstruction results.

[0021] The spatiotemporal graph convolutional network (ST-GCN) is used to analyze meteorological ground-based observation data and UAV detection data;

[0022] Optimize the physical rationality of meteorological fields provided by ground-based meteorological stations and UAVs through gradient constraint mechanisms;

[0023] New data is injected daily to update weights, allowing the LSTM-Transformer hybrid model to continuously learn. The new data includes ground-based meteorological observation data, ocean circulation data, drone detection data, and 3D wind field reconstruction results.

[0024] (4) Integrate meteorological ground-based observation data, ocean circulation data, and drone detection data to form a precise solution for near-surface micrometeorological conditions in complex terrain;

[0025] When vertical wind shear > 15m / s or turbulence intensity > 3m / s is detected, a risk heat map is generated;

[0026] Based on the LSTM-Transformer hybrid model, the system takes meteorological ground-based observation data and UAV detection data as input to automatically plan low-altitude flight paths to avoid strong convection areas.

[0027] Linked with the air traffic management system to push navigation warning information;

[0028] Support dynamic optimization of drone logistics routes;

[0029] (5) Achieve high-resolution meteorological monitoring and rapid forecasting and early warning, and provide precise navigation services for low-altitude airspace planning and control;

[0030] Dynamically identify severe convective areas based on the revised meteorological field and generate high-precision grid risk heat maps when vertical wind shear or turbulence intensity exceeds safety thresholds;

[0031] Combining the real-time meteorological field and digital elevation model provided by ground-based meteorological and drones, it automatically generates the optimal flight path to avoid dangerous areas;

[0032] Push risk warning and path planning results to the air traffic management system to trigger flight control instructions.

[0033] Beneficial effects of the present invention: The embodiment of the present invention uses a drone cluster equipped with Doppler lidar and microwave radiometer to achieve high-temporal and spatial resolution scanning of turbulence structures and temperature and humidity gradients in low altitudes below 1000 meters. The core technological breakthrough lies in the establishment of an air-sea-land intelligent coupling mechanism, which integrates meteorological ground-based observations, ocean circulation fields and drone detection data in real time based on the ensemble Kalman filter algorithm to build an efficient three-dimensional wind field reconstruction engine; and adopts a physically enhanced AI correction strategy, uses the LSTM-Transformer hybrid model to drive forecast updates, and introduces a gradient constraint mechanism to effectively suppress non-physical mutations in the meteorological field. The ultimate value of the system is reflected in the deep coordination of airspace control. It can automatically generate risk heat maps, plan obstacle avoidance paths, and link the air traffic management system to trigger control instructions. The comprehensive application of this system breaks through the limitations of traditional methods in temporal and spatial resolution, and provides high-temporal and spatial resolution early warning support for scenarios such as port scheduling and drone logistics. DETAILED DESCRIPTION

[0034] The specific implementation of the present invention is further described below in conjunction with the technical solution.

[0035] It is understandable that the separation of ocean circulation fields and atmospheric detection data in existing technologies leads to distortion in the simulation of sea-air interaction processes; at the same time, the high-dimensional heterogeneous data sent back by drones faces the lack of an efficient assimilation mechanism, and traditional Kalman filtering has insufficient performance when processing three-dimensional wind fields that are updated in seconds; in addition, the physical distortion problem of artificial intelligence meteorological models in complex terrain applications is significant, and its early warning results have not yet been effectively connected with the airspace control system.

[0036] To address these issues, the present invention proposes a Tianhai Zhimou weather perception system that uses drones to provide real-time corrections to airspace weather forecasts. This innovative approach leverages drones for real-time corrections. The system utilizes a swarm of drones equipped with Doppler lidar and microwave radiometers to achieve high-resolution spatial and temporal scanning of turbulence structures and temperature and humidity gradients below 1,000 meters. The core technological breakthrough lies in the establishment of an intelligent air-sea-land coupling mechanism. This system uses an ensemble Kalman filter algorithm to integrate ground-based meteorological observations, ocean circulation data, and drone detection data in real time to construct an efficient three-dimensional wind field reconstruction engine. Furthermore, the system employs a physically enhanced AI correction strategy, using a hybrid LSTM-Transformer model to drive forecast updates. Furthermore, a gradient constraint mechanism is introduced to effectively suppress non-physical fluctuations in the meteorological field. The ultimate value of this system lies in its deep collaboration with airspace management and control. It can automatically generate risk heat maps, plan obstacle avoidance paths, and trigger control commands in conjunction with the air traffic management system. The integrated application of this system overcomes the spatial and temporal resolution limitations of traditional methods, providing high-resolution early warning support for scenarios such as port scheduling and drone logistics.

[0037] A Tianhai Zhimou weather perception system for real-time correction of airspace weather forecast based on drones, the method comprising the following steps:

[0038] Step S101: construct a dynamic networking unit for a drone cluster. The drones are equipped with Doppler lidar, microwave radiometer and three-dimensional ultrasonic anemometer. Doppler lidar, microwave radiometer and three-dimensional ultrasonic anemometer are used to perform three-dimensional scanning of the low-altitude area and collect three-dimensional airspace micrometeorological data in real time.

[0039] In the embodiment of the present invention, a high load capacity ( 5kg) medium and large rotor UAVs, ensuring that they can carry multiple sensors; a Doppler lidar (wind field acquisition) is installed on the belly gimbal of the UAV; and an airborne microwave radiometer is deployed on the top of the fuselage, with an operating frequency band of 23.8GHz. 25MHz (for temperature and humidity inversion); three-dimensional ultrasonic anemometers (range 0-60m / s, sampling frequency 0-60m / s) are distributed at the end of the rotor arm. 10Hz); and configure an edge computing module to compress detection data in real time.

[0040] The construction of a dynamic networking unit for a drone swarm, equipped with a Doppler laser radar, a microwave radiometer, and a three-dimensional anemometer, for three-dimensional scanning of the low-altitude airspace and real-time collection of three-dimensional airspace micrometeorological data specifically includes the following steps:

[0041] Step S1011, select a device with high load capacity ( 5kg) medium and large rotor UAVs, ensuring that they can carry multiple sensors and install a Doppler lidar on the belly gimbal.

[0042] Step S1012: deploy the airborne microwave radiometer on the top of the fuselage, with the operating frequency band of 23.8 GHz. 25MHz (temperature and humidity inversion).

[0043] Step S1013: Install the three-dimensional ultrasonic anemometer at the end of the rotor arm (range 0-60m / s, sampling frequency 10Hz).

[0044] Step S1014: Use the edge computing module to compress the detection data in real time.

[0045] Furthermore, the Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecast by drones further includes the following steps:

[0046] Step S102, build a land-sea-air multi-source data fusion engine, and fuse meteorological ground-based observation data, ocean circulation field data and UAV detection data based on the ensemble Kalman filter algorithm; use the ensemble Kalman filter algorithm to fuse the meteorological ground-based observation data, ocean circulation field data and UAV detection data, and output the fusion three-dimensional wind field reconstruction result.

[0047] In the embodiment of the present invention, the meteorological ground-based observation data of the national ground stations are obtained in real time through the API protocol to realize the access of meteorological ground-based observation data. The meteorological ground-based observation data includes temperature, humidity, air pressure, precipitation, turbulence, and wind shear; the ocean circulation field input is obtained through the international open data platform (including but not limited to CMEMS, NASA PODAAC or NOAAOISST platform sea surface temperature and current data); aligning ground-based observation data, ocean circulation data, and drone detection data (including three-dimensional ultrasonic anemometer data, microwave radiometer-derived temperature and humidity, and Doppler lidar wind profiles) in time and space (based on a unified time and space benchmark for GPS / IMU data); using the ensemble Kalman filter algorithm to fuse multi-source observation data and assimilate atmospheric state variables (wind speed, temperature, humidity, precipitation, turbulence, wind shear, etc.); solving the fluid dynamics governing equations based on the assimilated high-resolution atmospheric state field and constructing a physically constrained three-dimensional variational wind field inversion model; performing vorticity correction and vertical velocity integration on the wind field to reconstruct a high-precision three-dimensional wind field structure; outputting the three-dimensional wind field reconstruction results (including U, V, W components and turbulence intensity information); and verifying the accuracy of the reconstructed wind field (horizontal wind speed RMSE) using meteorological tower ultrasonic anemometer data and wind profiler radar data obtained from the China Meteorological Administration. 1.5m / s, wind direction deviation 10°), and cross-validated with wind field data measured by offshore buoys in the offshore area.

[0048] The construction of a land-sea-air multi-source data fusion engine, which integrates meteorological ground-based observation station network, ocean circulation field and drone detection data based on the ensemble Kalman filter algorithm, specifically includes the following steps:

[0049] Step S1021, obtain meteorological ground-based observation data from ground stations across the country in real time through the API protocol, including temperature, humidity, air pressure, precipitation, turbulence, and wind shear data, to achieve access to meteorological ground-based observation data.

[0050] Step S1022: The ocean circulation field input is obtained through an international public data platform.

[0051] Step S1023: Using ensemble Kalman filtering to fuse meteorological ground-based observations, ocean circulation data, and UAV detection data.

[0052] Step S1024: output the three-dimensional wind field reconstruction result.

[0053] Furthermore, the Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecast by drones further includes the following steps:

[0054] In step S103, the LSTM-Transformer hybrid model is used to dynamically correct the three-dimensional wind field reconstruction result, and the accuracy of the LSTM-Transformer hybrid model is optimized in combination with the gradient constraint mechanism.

[0055] In an embodiment of the present invention, the input time series meteorological data (meteorological ground-based observation data, ocean circulation data, and three-dimensional wind field reconstruction results) is Z-score standardized and divided into time windows to construct batch data samples; an LSTM-Transformer hybrid model architecture is built: a long short-term memory network is used to extract long-term dependency features, and the final hidden state sequence of the long short-term memory network is input into the Transformer encoder; a spatiotemporal graph convolutional network is added to the output end of the Transformer encoder: meteorological elements are defined as graph nodes, a spatial adjacency matrix is ​​constructed using Euclidean distance and Gaussian kernel function, and the time convolution layer is combined to realize the spatiotemporal joint analysis of unstructured detection data (such as UAV distributed anemometer point cloud); in the LSTM-Transformer A physical constraint loss term is added to the output layer of the hybrid model: a gradient penalty function is constructed based on the fluid continuity equation and thermodynamic equations, and backpropagation is used to force the predicted field (wind speed u, v, w components, temperature T, humidity RH) to satisfy mass conservation and energy balance conditions. New data is injected at 00:00 UTC every day for incremental training, and an exponentially decaying learning rate strategy is used to update the weights of the LSTM-Transformer hybrid model for continuous learning. The European Centre for Medium-Range Weather Forecasts (ECMWF) analysis field is used as the benchmark truth value. The root mean square error (RMSE) and correlation coefficient (R²) of the predicted field within a 72-hour forecast window are compared to verify the generalization ability of the LSTM-Transformer hybrid model in extreme weather processes (typhoons and severe convection).

[0056] The dynamic correction of the three-dimensional wind field reconstruction results using the LSTM-Transformer hybrid model and the optimization of the LSTM-Transformer hybrid model accuracy in combination with the gradient constraint mechanism specifically include the following steps:

[0057] Step S1031: construct an LSTM-Transformer hybrid model to process time series meteorological data.

[0058] In step S1032, a spatiotemporal graph convolutional network (ST-GCN) is used to parse meteorological ground-based observation data and UAV detection data.

[0059] Step S1033: Optimize the physical rationality of the meteorological field provided by the meteorological base and the UAV through the gradient constraint mechanism.

[0060] In step S1034, new data is injected daily to update the weights so that the LSTM-Transformer hybrid model can continue to learn.

[0061] Furthermore, the Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecast by drones further includes the following steps:

[0062] Step S104: Integrate ground-based meteorological observation data, ocean circulation data, and drone detection data to form an accurate solution for near-surface micrometeorological conditions in complex terrain.

[0063] In an embodiment of the present invention, three-dimensional ultrasonic anemometer and Doppler lidar data are monitored in real time. When vertical wind shear > 15m / s or turbulence intensity > 3m / s is detected, the vertical wind shear and turbulence intensity data are processed using the Kriging interpolation algorithm to generate a high-resolution aviation risk heat map. Based on this heat map and the 72-hour weather forecast field (three-dimensional wind field, strong convection area), a search algorithm is used to automatically plan low-altitude flight paths and dynamically avoid high-risk areas (thermal map level). 6, severe convection area); push planned routes and warning information to the air traffic management system; support dynamic optimization of logistics drone routes.

[0064] The above-mentioned meteorological ground-based observation data, ocean circulation field data and drone detection data are used to form an accurate solution for near-surface micro-meteorological conditions in complex terrain, which specifically includes the following steps:

[0065] Step S1041 : When it is detected that the vertical wind shear is greater than 15 m / s or the turbulence intensity is greater than 3 m / s, a risk heat map is generated.

[0066] In step S1042, based on the LSTM-Transformer hybrid model, the meteorological ground-based observation data and the UAV detection data are used as input to automatically plan the low-altitude flight path to avoid the strong convection area.

[0067] Step S1043: Link with the air traffic management system to push navigation warning information.

[0068] Step S1044 supports dynamic optimization of drone logistics routes.

[0069] Furthermore, the Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecast by drones further includes the following steps:

[0070] Step S105: Implement high-resolution meteorological monitoring and rapid forecasting and warning, and provide precise navigation services for low-altitude airspace planning and control.

[0071] In an embodiment of the present invention, based on the three-dimensional meteorological field updated in real time by the ensemble Kalman filter, wind shear and turbulence mutations are detected online by airborne Doppler radar and three-dimensional ultrasonic anemometer; when the risk level is ≥7, the real-time navigation mechanism is triggered: the millimeter-wave radar point cloud and the terrain digital twin are integrated, and the artificial potential field method is used to dynamically generate a local obstacle avoidance vector; according to the vector instruction, the flight control system is controlled to adjust the attitude angle within 0.5 seconds to avoid sudden micro-downbursts; at the same time, the track is updated to the air traffic management system at high frequency through the data link to ensure synchronization of the airspace situation.

[0072] Among them, the realization of high-resolution meteorological monitoring and rapid forecasting and early warning provides precise navigation services for low-altitude airspace planning and control:

[0073] Step S1051: Dynamically identify severe convective areas based on the revised meteorological field, and generate a high-precision grid risk heat map when the vertical wind shear or turbulence intensity exceeds the safety threshold.

[0074] Step S1052 , combining the real-time meteorological field and digital elevation model provided by the meteorological ground base and the UAV, automatically generates an optimal flight path that avoids dangerous areas.

[0075] Step S1053: Push the risk warning and path planning results to the air traffic management system to trigger the flight control instructions.

[0076] The steps in the flowcharts of the various embodiments of the present invention are shown sequentially as indicated by the arrows, and therefore these steps need to be executed sequentially in the order indicated by the arrows. However, at least a portion of the steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages, especially the data collection portion, do not necessarily need to be completed at the same time, but may be executed at different times. The execution order of these sub-steps or stages does not necessarily need to be sequential, but may be executed in rotation or alternation with other steps or at least a portion of the sub-steps or stages of other steps.

[0077] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-described methods. The drones, lidars, edge computing units, multi-sensors, and mounted devices used in the various embodiments provided in this application are all rapidly iterating products, and the specific models are selected based on the user's budget and product iteration cycle.

[0078] The technical features of the above-mentioned embodiments cannot be combined arbitrarily and need to be carried out step by step in the order shown. However, as long as there are no fundamental differences between these technical features, they should be considered to be within the scope of this specification.

[0079] The above-described embodiments merely illustrate the implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A Tianhai Zhimou weather perception system based on real-time correction of airspace weather forecasts by drones, characterized by: Here are the steps: (1) Constructing a dynamic networking unit for drone swarms; The UAV swarm dynamic networking unit is equipped with a Doppler laser radar, microwave radiometer and three-dimensional ultrasonic anemometer on the drone. The Doppler laser radar, microwave radiometer and three-dimensional ultrasonic anemometer are used to perform three-dimensional scanning of the low-altitude area and collect three-dimensional airspace micro-meteorological data in real time. Use edge computing modules to compress three-dimensional airspace micrometeorological data in real time; The edge computing module includes a processor, memory and communication interface; (2) Build a land-sea-air multi-source data fusion engine to fuse meteorological ground-based observation data, ocean circulation field data, and UAV detection data based on the ensemble Kalman filter algorithm; The meteorological ground-based observation data, ocean circulation data and UAV detection data are fused using the ensemble Kalman filter algorithm to output the 3D wind field reconstruction result. (3) The LSTM-Transformer hybrid model is used to dynamically correct the 3D wind field reconstruction results, and the accuracy of the LSTM-Transformer hybrid model is optimized by combining the gradient constraint mechanism; (4) Integrate meteorological ground-based observation data, ocean circulation data, and drone detection data to form a precise solution for near-surface micrometeorological conditions in complex terrain; (5) Realize high-resolution meteorological monitoring and rapid forecasting and early warning, and provide precise navigation services for low-altitude airspace planning and control.

2. The Tianhai Intelligent Eyes weather perception system based on real-time correction of airspace weather forecast by drones according to claim 1 is characterized in that: In step (1), Choose a rotary-wing drone and ensure it carries multiple sensors; The Doppler lidar is installed on the belly gimbal of the drone; The microwave radiometer is deployed on top of the drone’s fuselage; The three-dimensional ultrasonic anemometer is distributedly installed at the end of the rotor arm of the UAV.

3. The Tianhai Intelligent Eyes weather perception system based on real-time correction of airspace weather forecast by drones according to claim 1 is characterized in that: In step (2), Real-time acquisition of meteorological ground-based observation data from ground stations across the country, including temperature, humidity, air pressure, precipitation, turbulence, and wind shear data. This data is then connected to the ensemble Kalman filter algorithm through the API protocol. Ocean circulation data are obtained through international open data platforms, including sea surface temperature and current data from CMEMS, NASA PODAAC, or NOAA OISST.

4. The Tianhai Intelligent Eyes weather perception system based on real-time correction of airspace weather forecast by drones according to claim 1 is characterized in that: In step (3), Build an LSTM-Transformer hybrid model to process time-series meteorological data. The LSTM-Transformer hybrid model includes a long short-term memory network, a self-attention mechanism, a spatiotemporal graph convolutional network, and a temporal convolutional layer. The self-attention mechanism includes an encoder and a decoder. The time-series meteorological data includes ground-based meteorological observation data, ocean circulation data, and three-dimensional wind field reconstruction results. Use spatiotemporal graph convolutional networks to analyze meteorological ground-based observation data and UAV detection data; Optimize the physical rationality of meteorological fields provided by ground-based meteorological stations and UAVs through gradient constraint mechanisms; New data is injected daily to update the weights, allowing the LSTM-Transformer hybrid model to continue learning; the new data includes meteorological ground-based observation data, ocean circulation field data, drone detection data, and three-dimensional wind field reconstruction results.

5. The Tianhai Intelligent Eyes weather perception system based on real-time correction of airspace weather forecast by drones according to claim 1 is characterized in that: In step (4), When vertical wind shear > 15m / s or turbulence intensity > 3m / s is detected, a risk heat map is generated; Based on the LSTM-Transformer hybrid model, the system takes meteorological ground-based observation data and UAV detection data as input to automatically plan low-altitude flight paths to avoid strong convection areas. Linked with the air traffic management system to push navigation warning information; Support dynamic optimization of drone logistics routes.

6. The Tianhai Intelligent Eyes weather perception system based on real-time correction of airspace weather forecast by drones according to claim 1 is characterized in that: In step (5), Dynamically identify severe convective areas based on the revised meteorological field and generate high-precision grid risk heat maps when vertical wind shear or turbulence intensity exceeds safety thresholds; Combining the real-time meteorological field and digital elevation model provided by ground-based meteorological and drones, it automatically generates the optimal flight path to avoid dangerous areas; Push risk warning and path planning results to the air traffic management system to trigger flight control instructions.