Unmanned aerial vehicle control method and device, electronic equipment and readable storage medium

Abstract

The invention provides an unmanned aerial vehicle control method and device, electronic equipment and a readable storage medium, and the method comprises the steps: obtaining multi-source heterogeneous meteorological data of a target airspace, and determining a space-time meteorological field of the target airspace based on the multi-source heterogeneous meteorological data; inputting the space-time meteorological field of the target airspace into the trained meteorological threat prediction model to obtain a threat prediction result of the target airspace; inputting the threat prediction result into a cross-scale risk transfer model to obtain a space-time risk thermodynamic diagram of the target unmanned aerial vehicle in the target airspace; and based on the space-time risk thermodynamic diagram, taking minimization of the space-time risk integral as an optimization target, determining a target flight path of the target unmanned aerial vehicle, and controlling the target unmanned aerial vehicle to fly according to the target flight path. Therefore, the space-time meteorological field generated by the multi-source heterogeneous meteorological data is converted into the ontology risk of the unmanned aerial vehicle, and the flight path is optimized in real time by taking the minimization of the space-time risk integral as the target, so that the control reliability of the unmanned aerial vehicle under the micrometeorological scale can be improved.

Description

Technical Field

[0001] The present application relates to the field of drone safety technology, and in particular to a drone control method, device, electronic device, and readable storage medium. Background Art

[0002] The use of unmanned aerial vehicles (UAVs) in low-altitude applications (generally defined as airspace below 100 meters above the ground) has seen explosive growth in numerous low-altitude sectors, including logistics and distribution, agricultural plant protection, power inspection, emergency response, and geographic mapping. However, the meteorological environment in low-altitude airspace is characterized by significant microscale (horizontal scales ranging from hundreds of meters to kilometers, and timescales ranging from minutes to tens of minutes), suddenness, strong localization, and dramatic variability. Wind shear (sharp changes in wind direction or speed over a short period of time), turbulence (irregular air movement), heavy rainfall (high-intensity precipitation affecting perception and propulsion), and low visibility (caused by fog, haze, smoke, and precipitation) are the four most prominent and frequent micrometeorological hazards posing a threat to UAV flight safety. These hazards are often difficult to effectively capture with traditional large-scale weather forecasts, yet can quickly lead to UAV attitude loss, navigation failure, structural damage, or even crashes, resulting in property damage and safety hazards. These hazards severely restrict the reliable operation and widespread use of UAVs in complex weather conditions.

[0003] Existing drone safety control generally relies on macro-meteorological forecast data (spatial resolution is mostly at the kilometer level, and the time update frequency is low) and the passive response mode of airborne sensors. It is difficult to meet the safety protection needs of drones for early perception, accurate assessment, and active avoidance of low-altitude micro-meteorological threats, resulting in poor control reliability of drones at the micro-meteorological scale. Summary of the Invention

[0004] In view of this, the embodiments of the present application at least provide a method, device, electronic device and readable storage medium for controlling a drone, which can improve the control reliability of the drone at the micrometeorological scale by converting the spatiotemporal meteorological field generated by multi-source heterogeneous meteorological data into the drone's ontological risk and optimizing the flight path in real time with the goal of minimizing the spatiotemporal risk integral.

[0005] This application mainly includes the following aspects: In a first aspect, an embodiment of the present application provides a method for controlling a drone, the method comprising: Acquiring multi-source heterogeneous meteorological data of a target airspace, and determining a spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data; the target airspace is the airspace where the planned flight path of the target UAV is located; Inputting the spatiotemporal meteorological field of the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; Inputting the threat prediction results into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace; Based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization goal, the target flight path of the target UAV is determined, and the target UAV is controlled to fly according to the target flight path.

[0006] In a second aspect, an embodiment of the present application further provides a control device for a drone, the control device for the drone comprising: a data fusion module configured to acquire multi-source heterogeneous meteorological data of a target airspace and determine, based on the multi-source heterogeneous meteorological data, a spatiotemporal meteorological field of the target airspace; the target airspace being the airspace where the planned flight path of the target UAV is located; A threat prediction module, configured to input the spatiotemporal meteorological field of the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; a risk transfer module, configured to input the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace; A control optimization module is used to determine the target flight path of the target UAV based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization goal, and control the target UAV to fly according to the target flight path.

[0007] In a third aspect, an embodiment of the present application further provides an electronic device comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the drone control method as described above.

[0008] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the drone control method described above are executed.

[0009] Embodiments of the present application provide a method, device, electronic device, and readable storage medium for controlling a drone. The method comprises: obtaining multi-source heterogeneous meteorological data for a target airspace, and determining a spatiotemporal meteorological field for the target airspace based on the multi-source heterogeneous meteorological data; the target airspace being the airspace within which the target drone's planned flight path lies; inputting the spatiotemporal meteorological field for the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; inputting the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map for the target drone in the target airspace; and determining a target flight path for the target drone based on the spatiotemporal risk heat map with minimizing the spatiotemporal risk integral as the optimization objective, and controlling the target drone to fly along the target flight path. In this way, by converting the spatiotemporal meteorological field generated by the multi-source heterogeneous meteorological data into the drone's intrinsic risk, and optimizing the flight path in real time with minimizing the spatiotemporal risk integral as the objective, the drone's control reliability at the micrometeorological scale can be improved.

[0010] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A flow chart of a method for controlling a drone provided in an embodiment of the present application is shown; Figure 2 One of the functional module diagrams of a drone control device provided in an embodiment of the present application is shown; Figure 3 A second functional module diagram of a drone control device provided in an embodiment of the present application is shown; Figure 4 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0014] When existing drones fly at low altitudes (especially below 100 meters), their safety assurance systems have significant technical defects and deficiencies when facing sudden, localized, micro-scale meteorological disasters (such as wind shear, turbulence, heavy rainfall, and low visibility). These defects are mainly manifested in the following aspects: Dynamic modeling of meteorological fields is lacking. Existing systems rely on static meteorological thresholds or isolated time series data, failing to capture the spatiotemporal coupled evolution of multi-scale meteorological fields. For example, the interaction between regional weather systems (such as typhoon circulation and high-pressure ridge movement) and microscale meteorology (such as local wind shear and sudden heavy rainfall cells) is not modeled. Meteorological data is processed at discrete time points or through simple linear extrapolation, ignoring the continuous dynamic changes in three-dimensional spatial structures (such as the migration paths of turbulent eddies and the diffusion trend of low-visibility fog clusters).

[0015] Threat predictions are fragmented in time and space. Airborne sensors only perceive the weather at the instantaneous location. Ground-based radar data (at the kilometer level) cannot resolve the threat core area at the hundred-meter level (such as the divergence point of a microburst). Traditional numerical model update frequency (at the hourly level) cannot match the minute-by-minute evolution of micrometeorological events, resulting in significant discrepancies between warnings and actual conditions.

[0016] Risk-based decision-making lacks cross-scale coordination. Existing strategies only respond to local, transient threats (such as single-point avoidance) and lack global path optimization based on regional meteorological evolution. For example, the large-scale wind shear belt triggered by the typhoon's outer circulation cannot be linked to the drone flight risk heat map; the trajectory of heavy rainfall cells moving with high-pressure systems is not incorporated into dynamic avoidance strategies.

[0017] It can be seen that the existing drone safety control generally relies on macro-meteorological forecast data (spatial resolution is mostly at the kilometer level, and the time update frequency is low) and the passive response mode of airborne sensors. It is difficult to meet the drone's safety protection needs for early perception, accurate assessment, and active avoidance of low-altitude micro-meteorological threats, resulting in poor control reliability of drones at the micro-meteorological scale.

[0018] In order to solve the above problems, the embodiments of the present application provide a control method, device, electronic device and readable storage medium for a drone. By converting the spatiotemporal meteorological field generated by multi-source heterogeneous meteorological data into the drone's ontological risk, and optimizing the flight path in real time with the goal of minimizing the spatiotemporal risk integral, the control reliability of the drone at the micrometeorological scale can be improved.

[0019] To facilitate understanding of the present application, the technical solutions provided in the present application are described in detail below in conjunction with specific embodiments.

[0020] See also Figure 1 , Figure 1 This is a flow chart of a method for controlling a drone provided in an embodiment of the present application. Figure 1 As shown, the control method of the drone provided in the embodiment of the present application includes the following steps: S101, acquiring multi-source heterogeneous meteorological data of a target airspace, and determining the spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data; the target airspace is the airspace where the planned flight path of the target UAV is located.

[0021] Here, the target airspace refers to the area covered by the target drone's planned flight path, representing the meteorological environment that requires key attention when the drone performs its mission. To comprehensively capture meteorological information in this area, the target drone acquires heterogeneous meteorological data from multiple sources. By fusing this multi-source, heterogeneous meteorological data, a spatiotemporal meteorological field for the target airspace is derived. This meteorological field dynamically reflects the spatiotemporal variations in meteorological parameters within the target airspace, providing fundamental data support for subsequent threat prediction and risk assessment. This process not only ensures data comprehensiveness but also lays a solid foundation for subsequent steps.

[0022] S102: Input the spatiotemporal meteorological field of the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result of the target airspace.

[0023] After acquiring the spatiotemporal weather field, the system inputs it into a trained meteorological threat prediction model. This model, a deep learning model, extracts spatial and temporal features from the spatiotemporal weather field and outputs threat predictions for the target airspace in the future. Specifically, the model extracts spatial features from the spatiotemporal weather field, learns the temporal evolution of these features, and ultimately outputs threat predictions for the future. In this way, the system can predict potential meteorological threats within the target airspace in advance, providing a crucial basis for subsequent risk assessment and path planning.

[0024] S103: Input the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace.

[0025] The threat prediction results are then fed into a cross-scale risk transfer model. This model converts the threat prediction results into the intrinsic risk of the target drone. Specifically, the model calculates the risk value corresponding to the threat prediction results and combines it with the drone's dynamic parameters (such as mass, airspeed, and moment of inertia) to generate a spatiotemporal risk heat map. This heat map covers the target airspace and evolves over time, intuitively reflecting the risk value of each location at different times. In this way, the system can quantify complex meteorological threats into intrinsic risks for the drone, providing a direct risk assessment basis for path planning.

[0026] S104: Based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization goal, determining a target flight path for the target UAV, and controlling the target UAV to fly according to the target flight path.

[0027] Here, the target flight path of the target UAV is determined based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization objective. The spatiotemporal risk integral is a quantitative metric used to assess the risk exposure of the UAV along its flight path. The optimal flight path is determined by searching the spatiotemporal risk heat map for a path that satisfies a preset risk threshold constraint and minimizes the risk integral. The target UAV is then controlled to fly along this path, ensuring it remains in a state of minimal risk throughout its flight. This approach achieves dynamic optimization and real-time control of the UAV's flight path, significantly improving the control reliability of the UAV under sudden micrometeorological threats.

[0028] Furthermore, the multi-source heterogeneous meteorological data includes ground sensor network data, air-based detection data, ground-based remote sensing data, and regional reanalysis data; and determining the spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data includes: Step a1: Time-align the ground sensor network data, air-based detection data, ground-based remote sensing data, and regional reanalysis data to obtain time-aligned multi-source heterogeneous meteorological data.

[0029] In this embodiment, the multi-source heterogeneous meteorological data covers data from various sources, including ground sensor network data, air-based detection data, ground-based remote sensing data, and regional reanalysis data.

[0030] Ground sensors acquiring ground sensor network data are deployed at a density of 50 nodes per square kilometer in the target airspace (e.g., 10 km x 10 km). These sensors use ultrasonic anemometers (with a measurement range of 0-30 m / s and an accuracy of ±0.1 m / s) and digital temperature and humidity sensors (with an accuracy of ±0.5°C / ±2% RH). This ground sensor network data is transmitted to an edge server via a low-power wide area network (LoRaWAN), which then sends it to the target drone with a transmission latency of ≤100 ms.

[0031] The airborne detection data is obtained by real-time calculation of the turbulent eddy dissipation rate (EDR) through the micro-weather station (sampling rate 10 Hz) and high-precision inertial navigation system (IMU) carried by the target UAV. The formula is: ; in is the airspeed (m / s), is the standard deviation of acceleration ( The data is then transmitted back to the target drone via 5G Ultra-Reliable Low-Latency Communication (URLLC), with an end-to-end latency of less than 5 ms.

[0032] Ground-based remote sensing data was acquired using a scanning Doppler lidar (1.5 μm wavelength) to obtain three-dimensional wind data. The scanning mode was an alternating planar position scan (PPI) and range-height scan (RHI) scan, with a spatial resolution of 30 m × 30 m and a data update rate of 1 Hz.

[0033] Regional reanalysis data uses the Global Forecast System (GFS) 0.25° grid data or the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data, which includes six key meteorological parameters: : Temperature at 2 meters height (℃); : surface relative humidity (%); : Surface solar radiation ( ); : total precipitation (mm); : Wind speed in east-west / north-south direction at 10 meters height (m / s).

[0034] This data comes from a wide range of sources and covers a variety of meteorological parameters, including wind speed, wind direction, temperature, humidity, and precipitation. To ensure data consistency and comparability, this data is first time-aligned. Time alignment involves aligning data from different sources to the same time base for subsequent processing. This ensures temporal consistency across all data, providing a foundation for subsequent spatial interpolation and fusion.

[0035] In the embodiment of the present application, the current time As a benchmark, a sliding time window is constructed. Historical data coverage time steps (5 minutes in length, total duration 6 hours), future forecast coverage In the next step (60 minutes), the multi-source heterogeneous meteorological data are time-aligned to obtain the time-aligned multi-source heterogeneous meteorological data.

[0036] Furthermore, the ground-based lidar that obtains ground-based remote sensing data can be replaced by the phased array weather radar (PAWR), which uses its strong rain and fog penetration ability to invert the wind field; the fixed sensor network can be expanded to the vehicle-mounted mobile platform, and the data can be dynamically compensated through V2X communication.

[0037] Step a2: uniformly map the time-aligned multi-source heterogeneous meteorological data to a spatial grid through a preset interpolation algorithm to obtain a four-dimensional tensor of the target airspace; the dimensions of the four-dimensional tensor are, in order, the historical time step dimension, the spatial grid row dimension, the spatial grid column dimension, and the meteorological parameter channel dimension.

[0038] Here, after completing the time alignment, these data are uniformly mapped to the spatial grid through a preset interpolation algorithm. The interpolation algorithm is used to fill discrete data points into a continuous spatial grid, thereby generating a complete spatiotemporal data field. In the embodiment of the present application, the target airspace is divided into The original GFS data (0.25°≈25 km) were refined to a 0.25 km grid using bilinear interpolation. Kriging was used to interpolate missing points in the ground sensor network. The semivariogram model is: ; in is the site spacing (meters), suitable for modeling the spatial correlation of wind speed, temperature, and humidity. Through interpolation, data of varying resolutions and coverage can be unified into a standard spatial grid, forming a four-dimensional tensor. The dimensions of this four-dimensional tensor are, in order, the historical time step dimension, the spatial grid row dimension, the spatial grid column dimension, and the meteorological parameter channel dimension. Specifically, the historical time step dimension represents the temporal sequence of the data, the spatial grid row and column dimensions represent the spatial distribution of the data, and the meteorological parameter channel dimension represents different meteorological parameters (such as wind speed, temperature, and humidity). In this way, the system can generate a spatiotemporal meteorological field that comprehensively reflects the meteorological conditions of the target airspace.

[0039] In the embodiment of the present application, the four-dimensional tensor can be expressed as Among them, the first dimension is the time step dimension, including 72 steps of historical data; the second and third dimensions constitute a 400×400 spatial grid; the fourth dimension is the meteorological parameter channel dimension, including meteorological parameters of six channels: temperature, humidity, solar radiation, precipitation, U wind (wind speed component in the east-west direction), and V wind (wind speed component in the north-south direction).

[0040] Step a3: determining the four-dimensional tensor of the target airspace as the space-time meteorological field of the target airspace.

[0041] Here, the generated four-dimensional tensor is defined as the spatiotemporal meteorological field of the target airspace. The spatiotemporal meteorological field is a dynamic data structure that reflects the spatiotemporal variations of meteorological parameters within the target airspace in real time. By fusing multi-source heterogeneous meteorological data into a unified spatiotemporal meteorological field, comprehensive and accurate data support can be provided for subsequent meteorological threat prediction and risk assessment. This process not only ensures data integrity and consistency but also provides a critical meteorological information foundation for drone flight safety.

[0042] Furthermore, the meteorological threat prediction model includes a spatial feature extraction layer, a temporal evolution layer, a first output layer, and a second output layer; inputting the spatiotemporal meteorological field of the target airspace into the trained meteorological threat prediction model to obtain a threat prediction result for the target airspace includes: Step b1: inputting the spatiotemporal meteorological field of the target airspace into the spatial feature extraction layer to extract the spatial meteorological structural features of the target airspace.

[0043] In this embodiment, the meteorological threat prediction model consists of multiple layers, the first of which is a spatial feature extraction layer. The function of this layer is to extract spatial meteorological structural features from the input spatiotemporal meteorological field. Specifically, the spatiotemporal meteorological field is a four-dimensional tensor that contains meteorological parameters at different time steps and spatial positions in the target airspace. Through the spatial feature extraction layer, the model can identify the spatial structure in the meteorological field, such as the shape of the wind shear core area, the distribution of turbulence, etc. In the embodiment of the present application, this process is implemented through a three-dimensional convolutional network (3D-CNN), which can capture the spatial correlation in the meteorological field and thus extract useful spatial features.

[0044] In the embodiment of the present application, the three-dimensional convolutional network includes a first convolutional layer, a pooling layer, and a second convolutional layer. The first convolutional layer uses a 3×3×3 convolution kernel (time × height × width), and the number of channels is increased from 6 to 64. This layer extracts local meteorological structural features, such as typhoon vortex closure isobars and wind shear divergence points. The pooling layer uses maximum pooling (2×2×2) to compress the feature map size and reduce computational complexity. The second convolutional layer uses a 3×3×3 convolution kernel and increases the number of channels to 128, capturing larger-scale meteorological patterns (such as the spatial distribution of frontal systems).

[0045] Step b2: inputting the spatial meteorological structural features of the target airspace into the temporal evolution layer to learn the temporal evolution features of the target airspace.

[0046] Here, after extracting the spatial meteorological structural features, these features are input into the time series evolution layer. The role of the time series evolution layer is to learn the changing patterns of these spatial features over time. Specifically, this layer can capture the temporal dynamic changes in the meteorological field, such as the moving path of the wind shear core area, the change in turbulence intensity, etc. In the embodiment of the present application, this process is implemented through a long short-term memory network (LSTM). LSTM can effectively process time series data and learn the temporal evolution patterns of meteorological features. In the embodiment of the present application, the feature sequence output by CNN is input into the LSTM unit in time steps, and the hidden layer dimension is 512. The update formula is: ; ; ; ; ; ; in, The output of the forget gate determines how much old information needs to be forgotten; Sigmoid activation function limits the output value to (0,1); is the weight matrix of the forget gate, which is used to calculate the activation value of the forget gate; is the hidden state of the previous time step, which contains the information of the previous time step; The spatial weather structure characteristics input for the current time step; is the bias term of the forget gate; The output of the input gate determines how much new information needs to be written; is the weight matrix of the input gate; is the bias term of the input gate; is the candidate memory, which is the new information that may be written in the current time step; It is a hyperbolic tangent activation function that limits the output value to (-1, 1); is the weight matrix of the candidate memory; is the bias term of the candidate memory; The memory cell state at the current time step stores long-term information; is the memory unit state at the previous time step; is element-wise multiplication (Hadamard product); The output of the output gate determines how much of the current memory needs to be output; is the weight matrix of the output gate; is the bias term of the output gate; is the hidden state of the spatial meteorological structural features of the current time step, which is used to be transferred to the next time step and as the output of the network; To perform a nonlinear transformation on the memory cell state so that its value is between (-1,1).

[0047] These formulas describe how an LSTM updates its internal state at each time step. The forget gate determines how much old information is forgotten, the input gate and candidate memory determine how much new information is written, the memory cell stores long-term information, and the output gate determines how much of the current memory is output. The hidden state contains this information and is passed to the next time step or as the network's output.

[0048] Step b3: input the time evolution characteristics of the target airspace into the first output layer and the second output layer respectively, and obtain the predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace respectively.

[0049] Here, after learning the time evolution features, these features are input into the first and second output layers, respectively. The first output layer is responsible for outputting the predicted wind shear trajectory, while the second output layer is responsible for outputting the predicted turbulence intensity field.

[0050] Specifically, the first output layer in the embodiment of the present application predicts the coordinates of the wind shear core area in the next 12 time steps through a fully connected network (FCN). and strength : , thus generating the predicted wind shear trajectory. is the wind shear intensity index (dimensionless, 0-1), which is calculated from the wind speed gradient Normalized to obtain.

[0051] The second output layer converts the LSTM hidden state into Upsample to the original spatial resolution and output the turbulent kinetic energy (TKE) distribution map: ; The unit is .Will Value > 35 Marked as heavy turbulence area (corresponding to civil aviation EDR>0.3).

[0052] In this way, the model can provide detailed meteorological threat information for subsequent risk assessment.

[0053] Step b4: determining the predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace as the threat prediction result of the target airspace.

[0054] Here, the model uses the predicted wind shear trajectory and turbulence intensity field as threat prediction results for the target airspace. These results include not only the movement path and intensity of the wind shear core area, but also the distribution of turbulence. In this way, the model can predict potential meteorological threats within the target airspace in advance, providing important basis for subsequent risk assessment and path planning.

[0055] In one possible implementation, the 3D-CNN+LSTM architecture can be replaced by a Transformer spatiotemporal model, which decomposes the meteorological tensor into spatiotemporal patches and captures the vortex migration patterns through a multi-head attention mechanism; or a physical information neural network (PINN) can be used to embed the residual term of the Navier-Stokes equation in the loss function to strengthen physical constraints.

[0056] Furthermore, the weather threat prediction model is trained according to the following steps: Step c1: constructing a training set sample based on the historical spatiotemporal meteorological field of the target airspace and the predicted wind shear trajectory and predicted turbulence intensity field corresponding to the historical spatiotemporal meteorological field.

[0057] In this embodiment, the first step in training the meteorological threat prediction model is to construct a training set of samples. Specifically, historical spatiotemporal meteorological data for the target airspace is collected. This data includes meteorological parameters such as wind speed, wind direction, temperature, humidity, and precipitation over a period of time. Simultaneously, the predicted wind shear trajectories and predicted turbulence intensity fields corresponding to these historical spatiotemporal meteorological fields are also collected. Together, these data constitute the training set of samples used to train the meteorological threat prediction model. For example, the spatiotemporal meteorological data for each day over the past year, along with the corresponding wind shear trajectories and turbulence intensity fields, can be collected to construct a training set rich in meteorological information.

[0058] Step c2: training the initial weather threat prediction model based on the training set samples.

[0059] Here, after constructing the training set samples, the initial meteorological threat prediction model is trained. In the embodiment of the present application, the initial meteorological threat prediction model is a deep learning model that includes a three-dimensional convolutional network (3D-CNN) and a long short-term memory network (LSTM) architecture. During training, the model gradually adjusts model parameters by learning the spatial and temporal features of the training set samples to improve the accuracy of predicting wind shear trajectories and turbulence intensity fields. The training process typically includes multiple iterative cycles, each of which updates the model parameters based on the training set samples to gradually improve the model's predictive performance.

[0060] Step c3: When the weighted sum of the wind shear trajectory loss function and the turbulence intensity field loss function is less than a preset threshold, a trained meteorological threat prediction model is generated.

[0061] Here, during the training process, the wind shear trajectory loss function and the turbulence intensity field loss function are used to evaluate the model's predictive performance. Specifically, the wind shear trajectory loss function is used to measure the difference between the wind shear trajectory predicted by the model and the actual wind shear trajectory, while the turbulence intensity field loss function is used to measure the difference between the turbulence intensity field predicted by the model and the actual turbulence intensity field. The weighted sum of these two loss functions is used as the overall loss function, and the model parameters are continuously optimized during the training process to minimize the overall loss function. When the overall loss function is less than the preset threshold, the system considers that the model has converged and generates a trained meteorological threat prediction model.

[0062] The loss function of the model training in the embodiment of the present application can be expressed as: ; in, is the coordinate of the true wind shear track; is the coordinate of the predicted wind shear track; is the total number of trajectory points; is the square of the Euclidean distance, which is used to calculate the difference between the predicted trajectory and the true trajectory; is the real turbulent kinetic energy (TKE) field; To predict the turbulent kinetic energy field, (Mean Squared Logarithmic Error) is the logarithmic mean square error, which is used to enhance the prediction accuracy of weak turbulence areas. It is particularly useful for processing data with a wide range, because it gives less weight to smaller errors and more weight to larger errors, specifically expressed as: . in, is the true value of the sample, is the predicted value of the sample, is the natural logarithm function.

[0063] Trajectory loss The performance of the model in predicting wind shear trajectories was evaluated by calculating the mean squared error (MSE) between the predicted and true trajectories. The smaller the trajectory loss, the closer the model-predicted wind shear trajectory is to the true trajectory.

[0064] Intensity field loss The model's performance in predicting turbulence intensity is evaluated by calculating the mean squared logarithmic error between the true and predicted turbulent kinetic energy fields. The smaller the intensity field loss, the closer the model's predicted turbulent kinetic energy field is to the true field.

[0065] In an embodiment of the present application, the trained weather threat prediction model is pruned and quantized and then deployed to the onboard computer of the target drone (such as NVIDIA Jetson AGX Orin), with an inference latency of ≤200 ms.

[0066] Furthermore, inputting the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace includes: Step d1: input the threat prediction result into the cross-scale risk transfer model to obtain the corresponding aircraft body risk of the target UAV in the target airspace.

[0067] In this embodiment, the threat prediction results (including the predicted wind shear trajectory and turbulence intensity field) are input into a cross-scale risk transfer model. This model quantifies the meteorological threat as a risk to the drone itself. Specifically, the model calculates wind shear risk and turbulence risk values, and combines them with the drone's dynamic parameters (such as mass, airspeed, and moment of inertia) to generate the drone's own risk.

[0068] For example, taking a typhoon as an example, the wind shear risk value can be calculated using the following formula: ; in, is the Euclidean distance between the typhoon center and the target UAV location (unit: km); is the distance attenuation; is the wind shear gradient, which serves as an intensity factor; is the terrain slope (normalized to 0-1), 0.7-1.0 in mountainous areas and 0-0.3 in plains; Terrain enhancement item.

[0069] Then, we can calculate the wind shear risk value based on the wind shear risk value. Constructing a posture instability probability model: ; in, is the mass of the drone (kg), is the airspeed (m / s), is the moment of inertia about the vertical axis ( ), is the stability coefficient (0.25 for quadrotor and 0.15 for fixed wing).

[0070] In this way, the model can convert complex meteorological threats into risk values ​​for the drone itself, providing a direct risk assessment basis for subsequent path planning.

[0071] Step d2: generating a spatiotemporal risk heat map covering the target airspace and evolving over time based on the aircraft body risk of the target UAV in the target airspace.

[0072] Here, after determining the drone's inherent risk, the system generates a spatiotemporal risk heat map based on these risk values, covering the target airspace and evolving over time. A spatiotemporal risk heat map is a dynamic risk map that reflects the risk value of each location within the target airspace at different times. Specifically, the system weights and fuses the wind shear risk value, turbulence risk value, and the risk values ​​of other threats (such as rainfall and visibility) to generate a comprehensive risk value. For example, the comprehensive risk value can be calculated using the following formula: ; in, is the wind shear risk value; is the turbulence risk value; is the rainfall risk value; is the visibility risk value; the weight can be adjusted dynamically according to the dominant threat. For example, when the rainfall intensity increases, The weight is increased to 0.6.

[0073] In addition, the risk propagation can be simulated by the heat conduction equation, so that the risk value can be smoothly propagated in the spatial grid. For example, the heat conduction equation can be expressed as: ; The explicit Euler method is used for discretization, and the time step is , spatial step length In this way, a spatiotemporal risk heat map that is updated over time can be generated, providing dynamic risk assessment for path planning.

[0074] Furthermore, the target flight path of the target UAV is determined based on the spatiotemporal risk heat map with minimization of the spatiotemporal risk integral as the optimization goal, including: In step e1, the spatiotemporal risk heat map is used as a cost base map to construct a risk accumulation index of the target UAV along the flight path.

[0075] In this example, a spatiotemporal risk heat map is used as the cost base map. This map covers the target airspace and evolves over time, intuitively reflecting the risk value of each location at different times. By integrating the risk values ​​in the spatiotemporal risk heat map along the flight path, a cumulative risk index for the target drone along its flight path is constructed.

[0076] Step e2: searching the spatiotemporal risk heat map for a flight path that satisfies a preset risk threshold constraint and minimizes the risk accumulation index, and obtaining a target flight path of the target UAV.

[0077] Here, after constructing the risk accumulation index, a flight path that satisfies the preset risk threshold constraint and minimizes the risk accumulation index is searched in the spatiotemporal risk heat map. Specifically, the embodiment of the present application adopts an improved A* algorithm, which combines the risk value and path length in the spatiotemporal risk heat map, evaluates the value of each node through a cost function, and thus selects the next node for expansion. By searching for a path that satisfies the preset risk threshold constraint, it is ensured that the risk value on the path does not exceed the safety upper limit. Ultimately, the system obtains the target flight path of the target drone, which can keep the drone in the lowest risk state at all times during flight.

[0078] In the embodiment of the present application, the cost function can be expressed as: ; in, is the average airspeed (m / s); is the target point coordinate. To accumulate risk :From the current step To Node The moment , the risk value corresponding to each grid point on the path With step length The products of are accumulated to get the total exposure risk. For inspiration :Use node With the target point The Euclidean distance divided by the average airspeed , and then multiply it by the weighting factor 1.5 to estimate the remaining time required for the voyage (risk exposure time).

[0079] therefore, , taking into account both the risks that have been taken and the risks of the parts that have not been taken, guiding the search to prioritize the expansion of paths with low cumulative risks and close to the target.

[0080] Among them, the preset risk threshold constraints include hard constraints: no entry Soft constraint: total risk integral of the path (Typical value 3.0).

[0081] Furthermore, model predictive control (MPC) can be used to track the drone’s path in real time and avoid risks. Specifically, a state space model is constructed: ; in, is the acceleration command ( ), is the risk field gradient. Solve the rolling optimization problem in seconds. Rolling optimization can be expressed as: ; Among them, the weight matrix ; .

[0082] In the embodiment of the present application, model predictive control (MPC) is used to perform real-time path tracking and risk avoidance on the UAV. Perform the following steps for the cycle: State update: Using discrete dynamics, the current position ,speed and risk field gradient Combined, predict the state of the next 10 steps. Rolling optimization: solve the quadratic programming, where Emphasis on position accuracy, Limit the acceleration amplitude, the coefficient of 10 ensures that the risk penalty is equivalent to the tracking error level. Instruction issued: only execute the optimization results , and then resamples the status and updates it in a rolling manner, achieving minute-level closed-loop obstacle avoidance and trajectory fine-tuning, significantly improving the flight safety margin in complex low-altitude meteorological environments.

[0083] Furthermore, the A*+MPC combination can be replaced by a deep reinforcement learning (DRL) strategy: with the risk field and drone state as input, the action space includes speed / altitude increments, and the reward function is directly mapped to the negative value of the cost function J.

[0084] Furthermore, the method further comprises: In step f1, if a flight path that meets the risk threshold constraint cannot be found in the spatiotemporal risk heat map, a safe landing area in the target airspace is determined based on a digital elevation model, a laser point cloud, and a Voronoi diagram partitioning algorithm.

[0085] In this embodiment, if the system fails to find a flight path that meets the risk threshold constraints within the spatiotemporal risk heat map, an emergency landing procedure is initiated. First, the system generates a terrain elevation map of the target airspace using a digital elevation model (DEM) and laser point cloud data. By analyzing the terrain geometry parameters in the elevation map, such as slope, roughness, and step depth, a landability analysis map is generated. Subsequently, a Voronoi diagram partitioning algorithm is used to partition the landable area and determine the safe landing zone. The Voronoi diagram divides the airspace into multiple subregions, where points within each subregion have the maximum distance to the nearest obstacle, thereby ensuring the safety of the landing area.

[0086] Step f2: searching for an emergency landing flight path that minimizes the risk accumulation index in the safe landing area by a gradient descent method to determine the optimal emergency landing point of the target UAV.

[0087] After determining safe landing areas, the optimal emergency landing path within these areas is searched for using the gradient descent method. Specifically, a risk accumulation index is used as the optimization objective function, and the gradient descent method is used to iteratively adjust the path parameters until the path that minimizes the risk accumulation index is found. The gradient descent method calculates the gradient of the objective function and adjusts the path parameters in the opposite direction of the gradient, gradually approaching the optimal solution. Ultimately, the optimal emergency landing point for the target drone is determined, which allows the drone to land with the lowest risk in an emergency.

[0088] In the embodiment of the present application, the optimization objective function of the optimal emergency landing path can be expressed as: ; in, For emergency landing paths; For the current moment; For landing time; For in time and path The integral represents the risk accumulation from the current moment to the landing moment; the optimization goal is to minimize the risk accumulation integral, that is, to find a path that minimizes the exposure of the target UAV to the risk area during flight. Constraints include slope constraint: the ground slope of the landing area must be less than 15 degrees. This is to ensure that the drone will not be in danger when landing due to excessive slope; obstacle density constraint: the obstacle density in the landing area must be less than 0.1 / This is to ensure that the landing area has enough clearance to prevent the drone from hitting obstacles during landing; Risk value constraint: at the moment of landing The risk value must be less than 0.3. This is to ensure that the drone is in a low-risk environment when landing.

[0089] Step f3: Control the target UAV to land at the optimal emergency landing point according to the emergency landing flight path.

[0090] Finally, the target drone is controlled to land at the optimal emergency landing point according to the determined emergency landing flight path. This process ensures that the drone can land safely and reliably in emergency situations, avoiding flight accidents caused by weather threats or other emergencies.

[0091] Furthermore, the embodiments of the present application can also adopt a federated learning architecture: the edge nodes locally cache data and aggregate them through a graph neural network (GNN) to generate a local meteorological field. After the CNN-LSTM model is federated trained, the DRL agent generates an obstacle avoidance strategy, retaining the core link of "meteorological modeling → risk transfer → decision-making".

[0092] An embodiment of the present application provides a method for controlling a drone, comprising: obtaining multi-source heterogeneous meteorological data for a target airspace, and determining a spatiotemporal meteorological field for the target airspace based on the multi-source heterogeneous meteorological data; the target airspace being the airspace within which the planned flight path of the target drone lies; inputting the spatiotemporal meteorological field for the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; inputting the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map for the target drone in the target airspace; and determining a target flight path for the target drone based on the spatiotemporal risk heat map with minimization of the spatiotemporal risk integral as the optimization objective, and controlling the target drone to fly along the target flight path. In this way, by converting the spatiotemporal meteorological field generated by the multi-source heterogeneous meteorological data into the drone's intrinsic risk, and optimizing the flight path in real time with minimization of the spatiotemporal risk integral as the objective, the drone's control reliability at the micrometeorological scale can be improved.

[0093] Based on the same application concept, the embodiments of the present application also provide a control device for a drone corresponding to the control method for a drone provided in the above embodiments. Since the principle of solving the problem by the device in the embodiments of the present application is similar to the control method for a drone in the above embodiments of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0094] See also Figure 2 , Figure 2 This is one of the functional module diagrams of a drone control device provided in an embodiment of the present application. Figure 2 As shown, the control device 200 of the drone provided in the embodiment of the present application includes: The data fusion module 210 is used to obtain multi-source heterogeneous meteorological data of the target airspace and determine the spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data; the target airspace is the airspace where the planned flight path of the target UAV is located.

[0095] The threat prediction module 220 is used to input the spatiotemporal meteorological field of the target airspace into the trained meteorological threat prediction model to obtain the threat prediction result of the target airspace.

[0096] The risk transfer module 230 is used to input the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace.

[0097] The control optimization module 240 is used to determine the target flight path of the target UAV based on the spatiotemporal risk heat map and to control the target UAV to fly according to the target flight path with minimizing the spatiotemporal risk integral as the optimization goal.

[0098] Furthermore, the multi-source heterogeneous meteorological data includes ground sensor network data, airborne detection data, ground-based remote sensing data, and regional reanalysis data; when the data fusion module 210 is used to determine the spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data, the data fusion module 210 is specifically used to: Time-aligning the ground sensor network data, airborne detection data, ground-based remote sensing data, and regional reanalysis data to obtain time-aligned multi-source heterogeneous meteorological data; The time-aligned multi-source heterogeneous meteorological data are uniformly mapped to a spatial grid through a preset interpolation algorithm to obtain a four-dimensional tensor of the target airspace; the dimensions of the four-dimensional tensor are the historical time step dimension, the spatial grid row dimension, the spatial grid column dimension, and the meteorological parameter channel dimension. The four-dimensional tensor of the target spatial domain is determined as the space-time meteorological field of the target spatial domain.

[0099] Furthermore, the meteorological threat prediction model includes a spatial feature extraction layer, a temporal evolution layer, a first output layer, and a second output layer. When the threat prediction module 220 inputs the spatiotemporal meteorological field of the target airspace into the trained meteorological threat prediction model to obtain a threat prediction result for the target airspace, the threat prediction module 220 is specifically configured to: Inputting the spatiotemporal meteorological field of the target airspace into the spatial feature extraction layer to extract the spatial meteorological structural features of the target airspace; Inputting the spatial meteorological structure characteristics of the target airspace into the temporal evolution layer to learn the temporal evolution characteristics of the target airspace; Inputting the time evolution characteristics of the target airspace into the first output layer and the second output layer respectively, to obtain the predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace respectively; The predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace are determined as the threat prediction result of the target airspace.

[0100] Furthermore, the threat prediction module 220 is further configured to train the weather threat prediction model according to the following steps: Constructing a training set sample based on the historical spatiotemporal meteorological field of the target airspace and the predicted wind shear trajectory and predicted turbulence intensity field corresponding to the historical spatiotemporal meteorological field; Training an initial meteorological threat prediction model based on the training set samples; When the weighted sum of the wind shear trajectory loss function and the turbulence intensity field loss function is less than a preset threshold, a trained meteorological threat prediction model is generated.

[0101] Furthermore, when the risk transfer module 230 is used to input the threat prediction result into the cross-scale risk transfer model to obtain the spatiotemporal risk heat map of the target UAV in the target airspace, the risk transfer module 230 is specifically used to: Inputting the threat prediction result into the cross-scale risk transfer model to obtain the corresponding aircraft body risk of the target UAV in the target airspace; Based on the aircraft body risk of the target UAV in the target airspace, a spatiotemporal risk heat map covering the target airspace and evolving over time is generated.

[0102] Furthermore, when the control optimization module 240 is used to determine the target flight path of the target UAV based on the spatiotemporal risk heat map and taking minimizing the spatiotemporal risk integral as the optimization goal, the control optimization module 240 is specifically used to: Using the spatiotemporal risk heat map as a cost base map, constructing a risk accumulation index for the target UAV along its flight path; A flight path that satisfies a preset risk threshold constraint and minimizes the risk accumulation index is searched in the spatiotemporal risk heat map to obtain a target flight path of the target UAV.

[0103] For further information, see Figure 3 , Figure 3 This is the second functional module diagram of a drone control device provided in an embodiment of the present application. Figure 3 As shown, the drone control device 200 provided in the embodiment of the present application further includes: The area judgment module 250 is used to determine the safe landing area in the target airspace based on the digital elevation model, laser point cloud and Voronoi diagram partitioning algorithm if a flight path that meets the risk threshold constraint cannot be found in the spatiotemporal risk heat map.

[0104] The landing planning module 260 is configured to search for an emergency landing flight path that minimizes the risk accumulation index in the safe landing area by using a gradient descent method, so as to determine an optimal emergency landing point for the target UAV.

[0105] The landing control module 270 is used to control the target UAV to land at the optimal emergency landing point according to the emergency landing flight path.

[0106] The present invention provides a control device for a drone, comprising: a data fusion module for acquiring multi-source heterogeneous meteorological data for a target airspace and, based on the multi-source heterogeneous meteorological data, determining a spatiotemporal meteorological field for the target airspace; the target airspace being the airspace within which the target drone's planned flight path lies; a threat prediction module for inputting the spatiotemporal meteorological field for the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; a risk transfer module for inputting the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map for the target drone in the target airspace; and a control optimization module for determining a target flight path for the target drone based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization objective, and controlling the target drone to fly along the target flight path. By converting the spatiotemporal meteorological field generated by the multi-source heterogeneous meteorological data into the drone's intrinsic risk and optimizing the flight path in real time with the objective of minimizing the spatiotemporal risk integral, the drone's control reliability at the micrometeorological scale can be improved.

[0107] Based on the same application idea, please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown, the electronic device 400 includes a processor 410 , a memory 420 and a bus 430 .

[0108] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 through the bus 430. When the processor 410 is running, the machine-readable instructions execute the steps of the drone control method provided in the above embodiment. The specific implementation method can be found in the method embodiment and will not be repeated here.

[0109] Based on the same application concept, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the drone control method provided in the above embodiment are executed. The specific implementation method can be found in the method embodiment and will not be repeated here.

[0110] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0111] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0112] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0113] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0114] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0115] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0116] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed in the present application, or make equivalent replacements for some of the technical features thereof. However, these modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for controlling a drone, characterized in that: The method comprises: Acquiring multi-source heterogeneous meteorological data of a target airspace, and determining a spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data; the target airspace is the airspace where the planned flight path of the target UAV is located; Inputting the spatiotemporal meteorological field of the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; Inputting the threat prediction results into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace; Based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization goal, the target flight path of the target UAV is determined, and the target UAV is controlled to fly according to the target flight path.

2. The method for controlling a drone according to claim 1, wherein: The multi-source heterogeneous meteorological data includes ground sensor network data, air-based detection data, ground-based remote sensing data and regional reanalysis data; The determining of the spatiotemporal meteorological field of the target airspace based on the multi-source heterogeneous meteorological data includes: Time-aligning the ground sensor network data, airborne detection data, ground-based remote sensing data, and regional reanalysis data to obtain time-aligned multi-source heterogeneous meteorological data; The time-aligned multi-source heterogeneous meteorological data are uniformly mapped to a spatial grid through a preset interpolation algorithm to obtain a four-dimensional tensor of the target airspace; the dimensions of the four-dimensional tensor are the historical time step dimension, the spatial grid row dimension, the spatial grid column dimension, and the meteorological parameter channel dimension. The four-dimensional tensor of the target spatial domain is determined as the space-time meteorological field of the target spatial domain.

3. The method for controlling a drone according to claim 1, wherein: The meteorological threat prediction model includes a spatial feature extraction layer, a temporal evolution layer, a first output layer, and a second output layer. Inputting the spatiotemporal meteorological field of the target airspace into the trained meteorological threat prediction model to obtain a threat prediction result for the target airspace includes: Inputting the spatiotemporal meteorological field of the target airspace into the spatial feature extraction layer to extract the spatial meteorological structural features of the target airspace; Inputting the spatial meteorological structure characteristics of the target airspace into the temporal evolution layer to learn the temporal evolution characteristics of the target airspace; Inputting the time evolution characteristics of the target airspace into the first output layer and the second output layer respectively, to obtain the predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace respectively; The predicted wind shear trajectory and the predicted turbulence intensity field of the target airspace are determined as the threat prediction result of the target airspace.

4. The method for controlling a drone according to claim 3, wherein: The weather threat prediction model is trained according to the following steps: Constructing a training set sample based on the historical spatiotemporal meteorological field of the target airspace and the predicted wind shear trajectory and predicted turbulence intensity field corresponding to the historical spatiotemporal meteorological field; Training an initial meteorological threat prediction model based on the training set samples; When the weighted sum of the wind shear trajectory loss function and the turbulence intensity field loss function is less than a preset threshold, a trained meteorological threat prediction model is generated.

5. The method for controlling a drone according to claim 1, wherein: Inputting the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace includes: Inputting the threat prediction result into the cross-scale risk transfer model to obtain the corresponding aircraft body risk of the target UAV in the target airspace; Based on the aircraft body risk of the target UAV in the target airspace, a spatiotemporal risk heat map covering the target airspace and evolving over time is generated.

6. The method for controlling a drone according to claim 1, wherein: The step of determining a target flight path of the target UAV based on the spatiotemporal risk heat map and taking minimizing the spatiotemporal risk integral as an optimization goal includes: Using the spatiotemporal risk heat map as a cost base map, constructing a risk accumulation index for the target UAV along its flight path; A flight path that satisfies a preset risk threshold constraint and minimizes the risk accumulation index is searched in the spatiotemporal risk heat map to obtain a target flight path of the target UAV.

7. The method for controlling a drone according to claim 6, wherein: The method further comprises: If a flight path that meets the risk threshold constraint cannot be found in the spatiotemporal risk heat map, a safe landing area in the target airspace is determined based on a digital elevation model, a laser point cloud, and a Voronoi diagram partitioning algorithm; Searching for an emergency landing flight path that minimizes the risk accumulation index in the safe landing area by a gradient descent method to determine an optimal emergency landing point for the target UAV; The target UAV is controlled to land at the optimal emergency landing point according to the emergency landing flight path.

8. A control device for a drone, characterized in that: The control device of the UAV includes: a data fusion module configured to acquire multi-source heterogeneous meteorological data of a target airspace and determine, based on the multi-source heterogeneous meteorological data, a spatiotemporal meteorological field of the target airspace; the target airspace being the airspace where the planned flight path of the target UAV is located; A threat prediction module, configured to input the spatiotemporal meteorological field of the target airspace into a trained meteorological threat prediction model to obtain a threat prediction result for the target airspace; a risk transfer module, configured to input the threat prediction result into a cross-scale risk transfer model to obtain a spatiotemporal risk heat map of the target UAV in the target airspace; A control optimization module is used to determine the target flight path of the target UAV based on the spatiotemporal risk heat map, with minimizing the spatiotemporal risk integral as the optimization goal, and control the target UAV to fly according to the target flight path.

9. An electronic device, characterized in that: include: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus, and the machine-readable instructions are executed by the processor to execute the steps of the control method of the drone according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for controlling a drone according to any one of claims 1 to 7 are executed.

Images