A method and system for controlling a drone considering weather risks

Abstract

The present application relates to a kind of unmanned aerial vehicle control method considering weather risk, including the following steps: real-time acquisition unmanned aerial vehicle flight environment data, synchronous receipt cloud layer evolution map and ground radar's wind shear early warning signal, process to obtain multimodal meteorological data;Predict the storm intensity distribution in future 2-5 minutes track area, and generate three-dimensional threat level grid map;For conventional meteorological risk, build threat level-avoidance strategy mapping table, judge threat level, and call corresponding avoidance strategy, guide unmanned aerial vehicle to execute corresponding avoidance action;When encountering extreme situation, activate anti-interference navigation mode, enhance flight stability by rotating speed-pitch angle collaborative control, trigger emergency protocol of alternate landing;After completing avoidance action, re-plan flight path in combination with space-time constraint, reasonably adjust flight speed.The present application also relates to a kind of unmanned aerial vehicle control system considering weather risk.The present application can improve the flight safety and stability of unmanned aerial vehicle under complex weather conditions.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a UAV control method and system taking weather risks into consideration. Background Art

[0002] In the modern logistics system, drone logistics services, with their high efficiency and flexibility, demonstrate tremendous development potential. However, the complex low-altitude flight environment, coupled with low-altitude wind shear and sudden weather changes, pose significant challenges to drone logistics. Low-altitude wind shear can rapidly alter a drone's flight speed and direction, leading to loss of control. Unexpected weather events, such as severe convection and heavy rain, can severely disrupt a drone's navigation and communication systems, even causing it to crash and resulting in cargo damage or loss.

[0003] Currently, traditional drones face numerous shortcomings when dealing with such complex weather conditions. For one thing, weather monitoring methods are limited. Relying on a small number of onboard sensors, it's difficult to obtain comprehensive and timely information about the surrounding weather, making it impossible to detect potentially dangerous weather conditions in advance. Furthermore, there's a lack of intelligent and sophisticated response strategies for severe weather. When faced with hazardous weather, drones are typically limited to simple hovering or return maneuvers, unable to flexibly adjust flight parameters based on the specific threat level. This significantly limits the reliability and application scope of drone logistics services, making it difficult to meet the growing demand for logistics and distribution. Summary of the Invention

[0004] In view of this, it is necessary to provide a UAV control method and system that takes weather risks into consideration.

[0005] The present invention provides a UAV control method considering weather risks, which comprises the following steps: a. real-time acquisition of UAV flight environment data by means of an onboard three-axis anemometer and a temperature and humidity sensor, synchronous reception of cloud evolution maps from meteorological satellites and 50-meter resolution wind shear warning signals from ground radars, and processing to obtain multimodal meteorological data; the UAV flight environment data comprises wind speed and direction, temperature and humidity data in the UAV flight environment; b. using a spatiotemporal convolutional neural network to fuse and analyze the multimodal meteorological data, predict the storm intensity distribution within the flight track area in the next 2-5 minutes, and generate a three-dimensional threat level grid map; c. for conventional meteorological wind d. When encountering extreme situations, the anti-interference navigation mode is automatically activated. Based on the quadrotor coupling dynamics model, the flight stability is enhanced through coordinated speed-pitch angle control, and the emergency landing protocol is triggered. e. After completing the evasive action in steps c and / or d, the flight path is replanned based on the improved genetic algorithm in combination with spatiotemporal constraints. The flight speed is reasonably adjusted through the speed increment allocation mechanism to eliminate the delay caused by the avoidance.

[0006] Preferably, the step a specifically includes:

[0007] Step S11, sensor installation and configuration: install the three-axis anemometer and temperature and humidity sensor at appropriate locations on the drone body to ensure that they can accurately sense external meteorological parameters;

[0008] Step S12, data transmission and processing: The data collected by the sensor and the data received by the communication module are transmitted in real time to the system data processing background via the wireless transmission module; the system data processing background performs preliminary screening and preprocessing on the data to remove abnormal data, providing a reliable data foundation for subsequent threat modeling;

[0009] Step S13, regular calibration and maintenance: send the sensor to a professional calibration agency for calibration regularly to ensure measurement accuracy; at the same time, check the signal strength and data transmission stability of the communication module, and replace aging or faulty components in a timely manner.

[0010] Preferably, the step b specifically includes:

[0011] Step S21, model training: Before the UAV is put into use, a large amount of meteorological data from different regions and weather conditions is collected, including historical meteorological data and simulated data; the spatiotemporal convolutional neural network is trained using this meteorological data, and the model parameters are adjusted to enable it to accurately predict the distribution of storm intensity;

[0012] Step S22, perform real-time prediction: When the drone is flying, the background processing server packages the real-time received meteorological monitoring data into a data tensor that conforms to the model input format and transmits it to the trained spatiotemporal convolutional neural network model in real time. The spatiotemporal convolutional neural network model outputs a three-dimensional threat level grid map and returns it to the hierarchical decision module.

[0013] Preferably, the threat level-avoidance strategy mapping table includes:

[0014] When the threat level is 1, dynamic heading compensation of ±5° is implemented to fine-tune the flight direction to cope with minor weather interference; when the threat level is 2, the altitude switching protocol is activated to avoid the dangerous area by changing the flight altitude; when the threat level is 3, a B-spline flight path with energy constraints is generated, thereby ensuring that the drone can safely avoid severe weather while reasonably controlling energy consumption.

[0015] Preferably, the step c specifically includes:

[0016] Step S31, establishing a mapping table: In the decision module, based on a large amount of experimental and actual flight data, a detailed threat level-avoidance strategy mapping table is established to clearly define specific operational instructions corresponding to different threat levels; the operational instructions include: heading adjustment angle, altitude layer switching range, and detour path generation rules;

[0017] Step S32, decision execution: After receiving the three-dimensional threat level grid map, the decision module quickly determines the threat level and calls the corresponding avoidance strategy from the mapping table. The strategy is transmitted to the flight control module through control instructions to guide the UAV to perform the corresponding avoidance action.

[0018] Preferably, the step d specifically includes:

[0019] Step S41, control algorithm implementation: In the flight control module, a precise speed-pitch angle coordinated control algorithm is written based on the quadrotor coupling dynamics model. This algorithm accurately adjusts the motor speed and pitch angle according to the preset route and real-time weather data to achieve high-precision track tracking.

[0020] Step S42, emergency mode triggering: When the sensor detects a continuous strong crosswind and the flight attitude is abnormal, the flight control module automatically activates the anti-interference navigation mode; in this mode, the control algorithm strengthens the adjustment of the motor speed and pitch angle to keep the drone flight stable; at the same time, the emergency landing protocol is triggered, and the alternate landing path is planned according to the surrounding geographical environment and the status of the drone, and an alarm is sent to the operator.

[0021] Preferably, the step e specifically includes:

[0022] Step S51, path replanning: After the avoidance action is completed, the mission efficiency compensation module is activated. The mission efficiency compensation module uses an improved genetic algorithm to replan the logistics path according to the spatiotemporal constraints to obtain the optimal path solution. The spatiotemporal constraints include the current location of the UAV, the remaining battery power, the destination, and the time limit.

[0023] Step S52, speed optimization: According to the newly planned route and the remaining delivery time, the flight speed of the drone in different sections is reasonably adjusted. If safety permits, the flight speed is appropriately increased to compensate for delivery delays and ensure that the logistics task is completed on time.

[0024] The present invention provides a UAV control system that takes weather risks into consideration. The system includes a meteorological data processing unit, a threat level map construction unit, an avoidance strategy decision unit, an extreme situation processing unit, and a mission effectiveness compensation unit. The meteorological data processing unit is used to collect UAV flight environment data in real time through an onboard three-axis anemometer and a temperature and humidity sensor, and synchronously receives cloud evolution maps from meteorological satellites and 50-meter resolution wind shear warning signals from ground radars to obtain multimodal meteorological data through processing. The UAV flight environment data includes wind speed and direction, temperature, and humidity data in the UAV flight environment. The threat level map construction unit is used to use a spatiotemporal convolutional neural network to perform fusion analysis on the multimodal meteorological data and predict the storm intensity distribution within the track area in the next 2-5 minutes. And generate a three-dimensional threat level grid map; the avoidance strategy decision unit is used to construct a threat level-avoidance strategy mapping table for conventional meteorological risks, judge the threat level according to the generated three-dimensional threat level grid map, and call the corresponding avoidance strategy from the threat level-avoidance strategy mapping table to guide the UAV to perform corresponding avoidance actions; the extreme situation processing unit is used to automatically activate the anti-interference navigation mode when encountering extreme situations, based on the four-rotor coupling dynamics model, through the speed-pitch angle coordinated control, enhance flight stability, and trigger the emergency landing protocol; the mission efficiency compensation unit is used to re-plan the flight path according to the improved genetic algorithm after completing the avoidance action, combined with time and space constraints, and reasonably adjust the flight speed through the speed increment distribution mechanism to eliminate the delay caused by avoidance.

[0025] The UAV control method and system of the present invention take weather risks into consideration, and improve the flight safety and stability of UAVs in complex weather conditions through multi-source meteorological perception, intelligent threat modeling, hierarchical decision-making, precise flight control, and mission effectiveness compensation, thereby ensuring the efficient completion of logistics and distribution tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the UAV control method considering weather risks of the present invention;

[0027] Figure 2 A schematic diagram of an implementation environment for a method for controlling a UAV taking weather risks into consideration provided by an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of the implementation process of step S2 provided in an embodiment of the present invention;

[0029] Figure 4 A schematic diagram of the implementation process of step S5 provided in an embodiment of the present invention;

[0030] Figure 5 This is a hardware architecture diagram of the UAV control system that takes weather risks into consideration according to the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] See Figure 1 The following is a flowchart of a preferred embodiment of the UAV control method considering weather risks of the present invention. Please refer to Figure 2 .

[0033] Step S1, multi-source meteorological sensing, is to collect UAV flight environment data in real time through an onboard three-axis anemometer and temperature and humidity sensor, and process it to obtain multimodal meteorological data. The UAV flight environment data includes wind speed and direction, temperature and humidity data in the UAV flight environment. Specifically:

[0034] The airborne triaxial anemometer measures the wind speed components in three orthogonal directions 、 、 , through the formula: , calculate the actual wind speed , the wind direction is based on 、 The ratio is calculated by the inverse tangent function. The temperature and humidity sensor collects the temperature and relative humidity Direct output. The onboard microcontroller reads the aforementioned sensor data and, utilizing a wireless communication module compliant with specific communication protocols (such as the IEEE 802.11p vehicle-to-vehicle communication protocol for communicating with the ground system's backend server), provides the system's backend program with real-time meteorological data collected during flight. The system's backend program receives, in real time, cloud evolution maps C from meteorological satellites (C transmitted in image format, containing information such as cloud top height, cloud amount, and cloud type) as well as 50-meter resolution wind shear warning signals W from ground radar (W encoded digitally, such as levels 0-5 representing varying wind shear risk levels). The data transmission rate meets real-time requirements to ensure timely access to comprehensive meteorological information.

[0035] Among them, for the real-time meteorological data sent back by the drone, the original data of wind speed, temperature and humidity are denoised and filtered through the sensor data preprocessing unit, and the Kalman filter algorithm is used. Its state equation is: ,in, The state vector of the system at time k is the core variable that needs to be estimated (i.e., key state parameters such as temperature, humidity, and wind speed in meteorological data); Represents the state transition matrix, which is used to describe the system from time At the time The law of state change (the evolution of meteorological state over time); Indicates that the system is at time The state vector of represents the control input matrix, which is used to describe the influence of external control variables on the system state (if it cannot be ignored, it can be set to 0); represents the external control input at time k-1 (such as the impact of manual intervention measures on local weather conditions); represents process noise, which is the uncertainty of the system model itself or the error caused by unmodeled factors (such as small disturbances that are not observed in the meteorological system). It is usually assumed to be Gaussian white noise; the observation equation is ,in, Indicates time The observation vector is the actual measurement data obtained from the sensor (such as wind speed, temperature, etc.); Represents the observation matrix, which is used to transform the system state vector Mapping to observation vector (Describes the mathematical relationship between states and observations, such as how a temperature state is converted into an electrical signal measurement from a sensor); is the observation noise, which obeys a Gaussian distribution with a mean of zero and a covariance of R. By continuously iteratively updating the state estimate, the noise interference is removed and more accurate meteorological parameter data is obtained.

[0036] Furthermore, the specific implementation method of this embodiment includes:

[0037] Step S11, sensor installation and configuration: install the three-axis anemometer and temperature and humidity sensor at appropriate locations on the drone body to ensure that they can accurately sense external meteorological parameters.

[0038] Use a high-precision three-axis anemometer, such as a certain brand's MEMS triaxial anemometer, with a measurement accuracy of ±0.1 m / s. Install it in a location on the drone body away from vibration sources such as the motor and secure it with a shock-absorbing bracket to ensure measurement accuracy. Use a digital temperature and humidity sensor, such as the SHT30, with a measurement accuracy of ±0.3°C for temperature and ±2 for relative humidity. Install it in a well-ventilated area of ​​the drone body with ample access to the outside air.

[0039] Step S12, data transmission and processing: The wind speed and direction, temperature and humidity data collected by the sensor are transmitted to the system's backend server in real time through the wireless communication module. The received data is preliminarily screened and preprocessed through the data processing program to remove abnormal data and save the data into the database to provide a reliable data foundation for subsequent threat modeling.

[0040] After receiving the sensor data transmitted by the communication module, the background data processing program unpacks and verifies it to ensure data integrity, and pre-processes the data. It uses a sliding average filter algorithm to remove high-frequency noise. For example, for wind speed data, the average value is calculated within a sliding window of length N. ,in, Indicates the data collected by the sensor Wind speed data The average value of Indicates the total number of representative data, that is, the number of wind speeds involved in the average calculation, which is a positive integer; Indicates the data samples, Counting starts at 0 and corresponds to different wind speed data in sequence. At the same time, for meteorological satellite cloud evolution maps and radar wind shear warning signals received from the meteorological department, multi-threading technology is used to improve data processing efficiency through sensor data processing threads, satellite data processing threads, and radar data processing threads.

[0041] Step S13, regular calibration and maintenance: send the sensor to a professional calibration agency for regular calibration to ensure measurement accuracy. At the same time, check the signal strength and data transmission stability of the communication module and replace aging or faulty components in a timely manner.

[0042] Every 50 flight hours, sensors are sent to a professional calibration laboratory for calibration using a standard weather simulator. The calibration equipment's accuracy is an order of magnitude higher than the sensor's measurement accuracy. The wireless communication module undergoes a monthly signal strength test using a signal strength meter to measure data transmission strength. If signal strength falls below a set threshold, the antenna connection and device aging issues are checked and components are replaced promptly.

[0043] Step S2, threat dynamic modeling: Use spatiotemporal convolutional neural networks to perform fusion analysis on multimodal meteorological data; by learning from historical meteorological data and real-time data, predict the storm intensity distribution within the track area in the next 2-5 minutes and generate a three-dimensional threat level grid map. Please refer to Figure 3 , specifically:

[0044] The input layer receives pre-processed multimodal meteorological data, including wind speed, wind direction, temperature, humidity, cloud evolution map feature vectors (extracted by image feature extraction algorithms such as VGG16), wind shear warning signals, etc. The convolution layer slides the convolution kernel in the spatiotemporal dimension to extract the spatiotemporal features in the data. The convolution operation formula is: ,in, Represents a single element of the output feature map, which means that after calculation, the coordinates in the output feature map are The value of the position is the result of the final calculation; Represents the double summation symbol, which traverses all elements of the convolution kernel (or weight matrix). 、 is the height and width of the convolution kernel (for example, Convolution kernel ), accumulate the calculation results of each position of the convolution kernel; : The local area of ​​the input feature map, representing the input feature map, with Base, offset The position element after is the original input fragment of the model (which can be understood as the local pixels of the image, the local features of the meteorological data, etc.); : The elements of the convolution kernel / weight matrix are the "parameters" obtained by model training, and the coordinates Corresponding to the position inside the convolution kernel, it is used to extract specific patterns of input features (such as edges, textures, meteorological patterns, etc.); : Bias term, which is the "offset" added to the calculation, is used to adjust the output results and make the model expression more flexible. The pooling layer uses the maximum pooling or average pooling method to reduce the data dimension and retain the main features. The fully connected layer maps the pooled feature vector to the final output dimension and outputs the probability distribution P(S) of the storm intensity at different locations in the track area in the next 2-5 minutes through the softmax function. The threat level is divided according to the probability distribution to generate a three-dimensional threat level grid map. The threat level L of each grid ranges from 1 to 3, where level 1 is low threat, level 2 is medium threat, and level 3 is high threat.

[0045] Furthermore, the specific implementation method of this embodiment includes:

[0046] Step S21: Model training: Before the drone is deployed, a large amount of meteorological data from different regions and weather conditions is collected, including historical meteorological data and simulated data. This data is used to train a spatiotemporal convolutional neural network and adjust the model parameters to accurately predict storm intensity distribution.

[0047] Meteorological data from different seasons and geographical regions (e.g., plains, mountainous areas, coastal areas, etc.) were collected, totaling N sets (N ≥ 10,000). Historical meteorological data was obtained from a meteorological database, and simulated data was generated using meteorological simulation software. The data was divided into a training set (80%), a validation set (10%), and a test set (10%). During training, the stochastic gradient descent (SGD) optimization algorithm was used with a learning rate of 0.001, a batch size of 32, and 500 training iterations. A spatiotemporal convolutional neural network model was constructed using the TensorFlow or PyTorch deep learning frameworks. The model training hardware platform used a workstation equipped with an NVIDIA GPU (e.g., RTX 3090) to accelerate the training process.

[0048] Step S22, perform real-time prediction: the background processing server packages the meteorological monitoring data received in real time into a data tensor that conforms to the model input format, and transmits it to the trained spatiotemporal convolutional neural network model in real time. The spatiotemporal convolutional neural network model outputs a three-dimensional threat level grid map and returns it to the hierarchical decision module to make flight adjustments for the UAV.

[0049] Step S3, hierarchical decision-making mechanism: For conventional meteorological risks, a threat level-avoidance strategy mapping table is constructed. The threat level is determined based on the generated three-dimensional threat level grid map, and the corresponding avoidance strategy is called from the threat level-avoidance strategy mapping table to guide the drone to perform the corresponding avoidance action:

[0050] At level 1 threat, the aircraft implements dynamic heading compensation of ±5° and fine-tunes the flight direction to cope with minor weather disturbances; at level 2 threat, the altitude switching protocol is activated to avoid dangerous areas by changing the flight altitude; at level 3 threat, a B-spline flight path with energy constraints is generated to ensure that the drone can safely avoid severe weather while reasonably controlling energy consumption. Specifically:

[0051] The threat level-avoidance strategy mapping table is as follows:

[0052] Level 1: Heading adjustment angle , the control instruction is ,in is the heading increment;

[0053] Level 2: Altitude switching range is , control instructions , the altitude switching is achieved by adjusting the throttle control signal of the UAV. According to the quadrotor dynamics model, the lift ( is the coefficient, is the motor speed), and the lift is adjusted by changing the motor speed to achieve height change;

[0054] Level 3: Generates a B-spline fly-around path with energy constraints. The mathematical expression of the B-spline curve is ,in is the B-spline basis function, The control point is determined by the optimization algorithm to ensure that the flight path meets the energy constraint conditions, such as total energy consumption. The power is a function of time variation, which does not exceed a certain proportion of the remaining energy of the drone (such as 70%). The control command .

[0055] Furthermore, the specific implementation method of this embodiment includes:

[0056] Step S31: Create a mapping table: In the decision module, a detailed threat level-avoidance strategy mapping table is established based on extensive experimental and actual flight data. This table clearly defines the specific operational instructions corresponding to different threat levels, such as heading adjustment angles, altitude switching ranges, and rules for generating circumvention paths.

[0057] Through extensive flight experiments, the drone's response performance was tested under various weather conditions, documenting the optimal avoidance strategies for different threat levels. At least 100 experiments were conducted, encompassing various wind speeds, directions, temperatures, humidity levels, and varying wind shear and storm conditions. Using this experimental data, combined with theoretical analysis and expert experience, a detailed and accurate mapping between threat level and avoidance strategy was established and stored in the decision module's non-volatile memory.

[0058] Step S32, Decision Execution: After receiving the 3D threat level grid map, the decision module quickly determines the threat level and calls the corresponding avoidance strategy from the mapping table. This strategy is communicated to the flight control module via control commands, instructing the drone to execute the corresponding avoidance maneuver.

[0059] The decision-making module, deployed on a backend server, utilizes a high-performance processor to rapidly process and make decisions. Upon receiving a three-dimensional threat level grid map, it determines the threat level within 10 milliseconds and retrieves the corresponding avoidance strategy from the mapping table. Control commands are sent to the drone's flight control module via a 50Hz PWM (pulse width modulation) signal, ensuring stable transmission of control commands.

[0060] Step S4: When encountering extreme situations, such as continuous strong crosswinds, the anti-interference navigation mode is automatically activated to reconfigure the flight control: that is, based on the quadrotor coupling dynamics model, the speed-pitch angle coordinated control is used to enhance flight stability and trigger the emergency landing protocol to ensure the safety of the drone in extreme situations. Specifically:

[0061] Based on the quadrotor coupling dynamics model, its dynamic equation is:

[0062]

[0063] in, : The total mass of the drone (including the fuselage, payload, etc., which determines the influence of inertia and gravity), : spatial position coordinates (describing the position of the aircraft in three-dimensional space, such as z corresponding to height), : Velocity component (the first derivative of position with respect to time, representing the velocity of the aircraft in direction of movement), : acceleration component (the first derivative of velocity with respect to time, determined by the force), : Gravity acceleration (constant vertical downward acceleration, affecting motion in the z direction), : Moment of inertia (describing the UAV's rotation The inertia of the shaft rotation is related to the structure and mass distribution). : attitude angle (corresponding to roll angle, pitch angle, and yaw angle, describing the spatial orientation of the drone), : is the angular velocity component (the first-order derivative of the attitude angle with respect to time, representing the speed of the drone's rotation around the axis), : angular acceleration component (the first derivative of angular velocity with respect to time, determined by the torque), : is the lift of the motor / rotor (the upward pull generated by the four rotors is the core power of the aircraft movement), : Air resistance coefficient (speed-related resistance that hinders the movement of the aircraft, such as The bigger, The greater the resistance), : Rotational damping coefficient (torque that hinders posture rotation, related to angular velocity, such as The bigger, The greater the damping), : is the geometric parameter ( Usually the distance from the rotor to the center of the fuselage, Related to the body structure and torque transmission), the above equation is divided into "translational dynamics" (the first 3 lines describe Position / velocity / acceleration of the direction) and "rotational dynamics" (the last 3 lines, describing The speed-pitch angle coordinated control algorithm adjusts the motor speed and pitch angle , so that the UAV flies along the preset route, and the control goal is to minimize the deviation between the actual track and the preset track , which is the Euclidean distance formula, used to calculate the straight-line distance between two points in three-dimensional space, where e: the Euclidean distance between the two points (the final calculation result, representing the straight-line length between the two points in space); : The spatial coordinates of the target point (here is the current position of the drone); : is the spatial coordinate of the reference point (or desired point) (here is the target waypoint of the UAV); the PID control algorithm is used, and the control law is ,in : is the controller output (the final calculation result, i.e. the motor speed of the drone), : is the proportional coefficient (adjusts the response strength of the "current error", the larger it is, the more sensitive it is to the error), : is the error at the current moment (the deviation between the actual position of the drone and the target position), : is the integral coefficient (adjusts the correction strength of "historical error accumulation" to eliminate long-term static errors), : is the differential coefficient (adjusts the response intensity of the "error change trend" and is used to predict the error), : is the integral term of the error (error About time The integral of , representing the “cumulative sum of historical errors”), : is the differential term of the error (error About time The first derivative of represents the "rate of error change." When encountering sustained strong crosswinds, the anti-interference navigation mode is activated, adding the anti-interference compensation term Uanti-wind to the control law to maintain flight stability. When the emergency landing protocol is activated, an alternate landing path is planned based on the surrounding geographic environment (e.g., terrain and buildings obtained through the Geographic Information System (GIS)) and the remaining battery power (Eremain), selecting the nearest and safest alternate landing location.

[0064] Furthermore, the specific implementation method of this embodiment includes:

[0065] Step S41: Control Algorithm Implementation: In the flight control module, a precise speed-pitch coordinated control algorithm is developed based on the quadrotor coupled dynamics model. This algorithm precisely adjusts the motor speed and pitch angle according to the preset flight path and real-time weather data, achieving high-precision flight path tracking.

[0066] The speed-pitch angle coordinated control algorithm and the anti-interference navigation mode algorithm are written in C on the flight control module's hardware platform (e.g., a flight control board based on the STM32H7 series microcontroller). The control algorithm's sampling frequency is set to 100Hz to track the drone's flight status in real time. A hardware timer generates a precise PWM signal with 12-bit resolution to control the motor speed. The drone's attitude angle is measured in real time using an inertial measurement unit (IMU). The IMU uses a six-axis or nine-axis sensor, such as the MPU9250, and is connected to the flight control board via the I2C bus.

[0067] Step S42, Emergency Mode Triggering: When sensors detect sustained strong crosswinds and abnormal flight attitude, the flight control module automatically activates anti-interference navigation mode. In this mode, the control algorithm enhances adjustments to motor speed and pitch angle to maintain stable flight. Simultaneously, the emergency landing protocol is triggered, planning an alternate landing path based on the surrounding geographical environment and the drone's status, and alerting the operator.

[0068] The sensor detects a continuous strong crosswind when the crosswind speed Vside exceeds the set threshold (e.g. 10m / s) and the duration exceeds the critical value t (e.g. 5 seconds). When the conditions are met, the flight control module activates the anti-interference navigation mode within 100 milliseconds and sends an alert message to the operator through the wireless communication module (e.g. 4G module), including the drone's location, remaining battery power, current weather conditions, etc. The alternate landing path planning algorithm uses The algorithm, combined with geographic information system data, plans the optimal alternate landing path within 5 seconds.

[0069] Step S5: Mission performance compensation: After completing the avoidance actions in step S3 and / or step S4, the flight path is replanned based on the improved genetic algorithm and combined with time and space constraints; the flight speed is reasonably adjusted through the speed increment allocation mechanism to eliminate the delay caused by avoidance and ensure on-time delivery. Figure 4 , specifically:

[0070] The initial population consists of randomly generated logistics path chromosomes. Each chromosome encodes a possible logistics path, and the path encoding method uses a node sequence representation. The fitness function f comprehensively considers factors such as path length L, delivery time T, and remaining power E. For example, ,in 、 、 The population is continuously evolved through selection, crossover, and mutation operations. The selection operation uses the roulette wheel selection method, the crossover operation uses the partial matching crossover (PMX) or sequential crossover (OX) method, and the mutation operation uses the exchange mutation or insertion mutation method. After multiple iterations, the optimal logistics path that meets the time and space constraints is found. The speed increment allocation mechanism calculates the speed increment of each section based on the newly planned path and the remaining delivery time Tremain. , the formula is ,in is the road section distance, Adjust the drone's flight speed within the originally planned speed and safety limits to eliminate delivery delays.

[0071] Furthermore, the specific implementation method of this embodiment includes:

[0072] Step S51, Path Replanning: After the avoidance maneuver is completed, the mission efficiency compensation module is activated. This module uses a modified genetic algorithm to replan the logistics route based on spatiotemporal constraints such as the drone's current location, remaining battery power, destination, and time limit. Through multiple iterative calculations, the optimal path solution is found.

[0073] The mission effectiveness compensation module is based on an improved genetic algorithm, with an initial population size of 50 and an evolutionary number of 100. Geographic Information System (GIS) data is used to obtain information about the drone's current location, destination, and surrounding geographic environment, such as obstacle locations and no-fly zones. This information is incorporated into the fitness function as a constraint to ensure the planned path is safe and feasible.

[0074] Step S52, Speed ​​Optimization: Based on the newly planned route and remaining delivery time, the mission efficiency compensation module uses a speed increment allocation mechanism to rationally adjust the drone's flight speed along different sections of the route. Where safety permits, the flight speed is appropriately increased to compensate for delivery delays and ensure the logistics mission is completed on time.

[0075] The speed increment allocation mechanism is activated after the path replanning is completed. The speed increment of each section is calculated based on the new planned path and the remaining delivery time. In practical applications, considering the power performance and flight safety of the UAV, the speed increment upper limit is set to By adjusting the drone's throttle control signal, speed optimization is achieved, minimizing delivery delays while ensuring flight safety.

[0076] See Figure 5 Figure 1 shows the hardware architecture of a weather-risk-based UAV control system 10 according to the present invention. The system includes a meteorological data processing unit 101, a threat level map construction unit 102, an avoidance strategy decision unit 103, an extreme situation processing unit 104, and a mission effectiveness compensation unit 105.

[0077] The meteorological data processing unit 101 is used to perform multi-source meteorological sensing, that is, to collect drone flight environment data in real time through an onboard three-axis anemometer and temperature and humidity sensor, and process it to obtain multi-modal meteorological data. The drone flight environment data includes: wind speed and direction, temperature and humidity data in the drone flight environment. Specifically:

[0078] The airborne triaxial anemometer measures the wind speed components in three orthogonal directions 、 、 , through the formula: , calculate the actual wind speed , the wind direction is based on 、 The ratio is calculated by the inverse tangent function. The temperature and humidity sensor collects the temperature and relative humidity Direct output. The aforementioned sensor data is read by an onboard microcontroller, and a wireless communication module conforming to a specific communication protocol (such as the IEEE802.11p vehicle-to-vehicle communication protocol for communicating with the ground system backend server) is used to provide the system backend program with real-time meteorological data collected during flight. The system backend program receives, in real time, cloud evolution maps C from meteorological satellites (C is transmitted in image format and contains information such as cloud top height, cloud amount, and cloud type) as well as 50-meter resolution wind shear warning signals W from ground radar (W is digitally encoded, such as levels 0-5 representing different levels of wind shear risk). The data transmission rate must meet real-time requirements to ensure timely access to comprehensive meteorological information.

[0079] Among them, for the real-time meteorological data sent back by the drone, the original data of wind speed, temperature and humidity are denoised and filtered through the sensor data preprocessing unit, and the Kalman filter algorithm is used. Its state equation is: ,in, The state vector of the system at time k is the core variable that needs to be estimated (i.e., key state parameters such as temperature, humidity, and wind speed in meteorological data); Represents the state transition matrix, which is used to describe the system from time At the time The law of state change (the evolution of meteorological state over time); Indicates that the system is at time The state vector of represents the control input matrix, which is used to describe the influence of external control variables on the system state (if it cannot be ignored, it can be set to 0); represents the external control input at time k-1 (such as the impact of manual intervention measures on local weather conditions); represents process noise, which is the uncertainty of the system model itself or the error caused by unmodeled factors (such as small disturbances that are not observed in the meteorological system). It is usually assumed to be Gaussian white noise; the observation equation is ,in, Indicates time The observation vector is the actual measurement data obtained from the sensor (such as wind speed, temperature, etc.); Represents the observation matrix, which is used to transform the system state vector Mapping to observation vector (Describes the mathematical relationship between states and observations, such as how a temperature state is converted into an electrical signal measurement from a sensor); is the observation noise, which obeys a Gaussian distribution with a mean of zero and a covariance of R. By continuously iteratively updating the state estimate, the noise interference is removed and more accurate meteorological parameter data is obtained.

[0080] Furthermore, the specific implementation of this embodiment includes:

[0081] Sensor installation and configuration: Install the three-axis anemometer and temperature and humidity sensors at appropriate locations on the drone body to ensure that they can accurately sense external meteorological parameters.

[0082] Use a high-precision three-axis anemometer, such as a certain brand's MEMS triaxial anemometer, with a measurement accuracy of ±0.1 m / s. Install it in a location on the drone body away from vibration sources such as the motor and secure it with a shock-absorbing bracket to ensure measurement accuracy. Use a digital temperature and humidity sensor, such as the SHT30, with a measurement accuracy of ±0.3°C for temperature and ±2 for relative humidity. Install it in a well-ventilated area of ​​the drone body with ample access to the outside air.

[0083] Data transmission and processing: The data collected by the sensors and the data received by the communication module are transmitted to the system data processing background in real time through the wireless transmission module; the system background performs preliminary screening and preprocessing on the data, removes abnormal data, and provides a reliable data foundation for subsequent threat modeling.

[0084] After receiving the sensor data transmitted by the communication module, the background data processing program unpacks and verifies it to ensure data integrity, and pre-processes the data. It uses a sliding average filter algorithm to remove high-frequency noise. For example, for wind speed data, the average value is calculated within a sliding window of length N. ,in, Indicates the data collected by the sensor Wind speed data The average value of Indicates the total number of representative data, that is, the number of wind speeds involved in the average calculation, which is a positive integer; Indicates the data samples, Counting starts at 0 and corresponds to different wind speed data in sequence. At the same time, for meteorological satellite cloud evolution maps and radar wind shear warning signals received from the meteorological department, multi-threading technology is used to improve data processing efficiency through sensor data processing threads, satellite data processing threads, and radar data processing threads.

[0085] Regular calibration and maintenance: Send the sensor to a professional calibration agency for regular calibration to ensure measurement accuracy. Also, check the signal strength and data transmission stability of the communication module and replace aging or faulty components in a timely manner.

[0086] Every 50 flight hours, sensors are sent to a professional calibration laboratory for calibration using a standard weather simulator. The calibration equipment's accuracy is an order of magnitude higher than the sensor's measurement accuracy. The communication module undergoes a monthly signal strength test using a signal strength meter to measure data transmission strength. If signal strength falls below a set threshold, the antenna connection and equipment aging issues are checked and components are replaced promptly.

[0087] The threat level map construction unit 102 is used to perform dynamic threat modeling: using a spatiotemporal convolutional neural network to perform fusion analysis on multimodal meteorological data; by learning from historical meteorological data and real-time data, it predicts the storm intensity distribution within the track area in the next 2-5 minutes and generates a three-dimensional threat level grid map. Figure 3 , specifically:

[0088] The input layer receives pre-processed multimodal meteorological data, including wind speed, wind direction, temperature, humidity, cloud evolution map feature vectors (extracted by image feature extraction algorithms such as VGG16), wind shear warning signals, etc. The convolution layer slides the convolution kernel in the spatiotemporal dimension to extract the spatiotemporal features in the data. The convolution operation formula is: ,in, Represents a single element of the output feature map, which means that after calculation, the coordinates in the output feature map are The value of the position is the result of the final calculation; Represents the double summation symbol, which traverses all elements of the convolution kernel (or weight matrix). 、 is the height and width of the convolution kernel (e.g. Convolution kernel ), accumulate the calculation results of each position of the convolution kernel; : The local area of ​​the input feature map, representing the input feature map, with Base, offset The position element after is the original input fragment of the model (which can be understood as the local pixels of the image, the local features of the meteorological data, etc.); : The elements of the convolution kernel / weight matrix are the "parameters" obtained by model training, and the coordinates Corresponding to the position inside the convolution kernel, it is used to extract specific patterns of input features (such as edges, textures, meteorological patterns, etc.); : Bias term, which is the "offset" added to the calculation, is used to adjust the output results and make the model expression more flexible. The pooling layer uses the maximum pooling or average pooling method to reduce the data dimension and retain the main features. The fully connected layer maps the pooled feature vector to the final output dimension and outputs the probability distribution P(S) of the storm intensity at different locations in the track area in the next 2-5 minutes through the softmax function. The threat level is divided according to the probability distribution to generate a three-dimensional threat level grid map. The threat level L of each grid ranges from 1 to 3, where level 1 is low threat, level 2 is medium threat, and level 3 is high threat.

[0089] Furthermore, the specific implementation of this embodiment includes:

[0090] The threat level map construction unit 102 performs model training. Before the drone is deployed, it collects a large amount of meteorological data from different regions and weather conditions, including historical meteorological data and simulated data. This data is used to train a spatiotemporal convolutional neural network and adjust model parameters to accurately predict storm intensity distribution.

[0091] Meteorological data from different seasons and geographical regions (e.g., plains, mountainous areas, coastal areas, etc.) were collected, totaling N sets (N ≥ 10,000). Historical meteorological data was obtained from a meteorological database, and simulated data was generated using meteorological simulation software. The data was divided into a training set (80%), a validation set (10%), and a test set (10%). During training, the stochastic gradient descent (SGD) optimization algorithm was used with a learning rate of 0.001, a batch size of 32, and 500 training iterations. A spatiotemporal convolutional neural network model was constructed using the TensorFlow or PyTorch deep learning frameworks. The model training hardware platform used a workstation equipped with an NVIDIA GPU (e.g., RTX 3090) to accelerate the training process.

[0092] The threat level map construction unit 102 performs real-time prediction: when the UAV is flying, the background processing server packages the meteorological monitoring data received in real time into a data tensor that conforms to the model input format, and transmits it to the trained spatiotemporal convolutional neural network model in real time. The spatiotemporal convolutional neural network model outputs a three-dimensional threat level grid map and returns it to the hierarchical decision module to make flight adjustments for the UAV.

[0093] The avoidance strategy decision unit 103 is used to implement a hierarchical decision mechanism: for conventional meteorological risks, a threat level-avoidance strategy mapping table is constructed, the threat level is determined based on the generated three-dimensional threat level grid map, and the corresponding avoidance strategy is called from the threat level-avoidance strategy mapping table to guide the UAV to perform the corresponding avoidance action:

[0094] At level 1 threat, the aircraft implements dynamic heading compensation of ±5° and fine-tunes the flight direction to cope with minor weather disturbances; at level 2 threat, the altitude switching protocol is activated to avoid dangerous areas by changing the flight altitude; at level 3 threat, a B-spline flight path with energy constraints is generated to ensure that the drone can safely avoid severe weather while reasonably controlling energy consumption. Specifically:

[0095] The threat level-avoidance strategy mapping table is as follows:

[0096] Level 1: Heading adjustment angle , the control instruction is ,in is the heading increment;

[0097] Level 2: Altitude switching range is , control instructions , the altitude switching is achieved by adjusting the throttle control signal of the UAV. According to the quadrotor dynamics model, the lift ( is the coefficient, is the motor speed), and the lift is adjusted by changing the motor speed to achieve height change;

[0098] Level 3: Generates a B-spline fly-around path with energy constraints. The mathematical expression of the B-spline curve is ,in is the B-spline basis function, The control point is determined by the optimization algorithm to ensure that the flight path meets the energy constraint conditions, such as total energy consumption. The power is a function of time variation, which does not exceed a certain proportion of the remaining energy of the drone (such as 70%). The control command .

[0099] Furthermore, the specific implementation of this embodiment includes:

[0100] The avoidance strategy decision unit 103 establishes a mapping table: In the decision module, a detailed threat level-avoidance strategy mapping table is established based on a large amount of experimental and actual flight data. The specific operational instructions corresponding to different threat levels are clearly defined, such as heading adjustment angles, altitude layer switching ranges, and rules for generating circumvention paths.

[0101] Through extensive flight experiments, the drone's response performance was tested under various weather conditions, documenting the optimal avoidance strategies for different threat levels. At least 100 experiments were conducted, encompassing various wind speeds, directions, temperatures, humidity levels, and varying wind shear and storm conditions. Using this experimental data, combined with theoretical analysis and expert experience, a detailed and accurate mapping between threat level and avoidance strategy was established and stored in the decision module's non-volatile memory.

[0102] The avoidance strategy decision unit 103 executes the decision: upon receiving the three-dimensional threat level grid map, the decision module quickly determines the threat level and calls the corresponding avoidance strategy from the mapping table. This strategy is then communicated to the flight control module via control instructions, instructing the drone to execute the corresponding avoidance maneuver.

[0103] The decision-making module, deployed on a backend server, utilizes a high-performance processor to rapidly process and make decisions. Upon receiving a three-dimensional threat level grid map, it determines the threat level within 10 milliseconds and retrieves the corresponding avoidance strategy from the mapping table. Control commands are sent to the drone's flight control module via a 50Hz PWM (pulse width modulation) signal, ensuring stable transmission of control commands.

[0104] The extreme situation processing unit 104 is used to automatically activate the anti-interference navigation mode and perform flight control reconstruction when encountering extreme situations, such as continuous strong crosswinds: that is, based on the quadrotor coupling dynamics model, through speed-pitch angle coordinated control, enhance flight stability and trigger the emergency landing protocol to ensure the safety of the UAV in extreme situations. Specifically:

[0105] Based on the quadrotor coupling dynamics model, its dynamic equation is:

[0106]

[0107] in : The total mass of the drone (including the fuselage, payload, etc., which determines the influence of inertia and gravity), : spatial position coordinates (describing the position of the aircraft in three-dimensional space, such as z corresponding to height), : Velocity component (the first derivative of position with respect to time, representing the velocity of the aircraft in direction of movement), : acceleration component (the first derivative of velocity with respect to time, determined by the force), : Gravity acceleration (constant vertical downward acceleration, affecting motion in the z direction), : Moment of inertia (describing the UAV's rotation The inertia of the shaft rotation is related to the structure and mass distribution). : attitude angle (corresponding to roll angle, pitch angle, and yaw angle, describing the spatial orientation of the drone), : is the angular velocity component (the first-order derivative of the attitude angle with respect to time, representing the speed of the drone's rotation around the axis), : angular acceleration component (the first derivative of angular velocity with respect to time, determined by the torque), : is the lift of the motor / rotor (the upward pull generated by the four rotors is the core power of the aircraft movement), : Air resistance coefficient (speed-related resistance that hinders the movement of the aircraft, such as The bigger, The greater the resistance), : Rotational damping coefficient (torque that hinders posture rotation, related to angular velocity, such as The bigger, The greater the damping), : is the geometric parameter ( Usually the distance from the rotor to the center of the fuselage, Related to the body structure and torque transmission), the above equation is divided into "translational dynamics" (the first 3 lines describe Position / velocity / acceleration of the direction) and "rotational dynamics" (the last 3 lines, describing The speed-pitch angle coordinated control algorithm adjusts the motor speed and pitch angle , so that the UAV flies along the preset route, and the control goal is to minimize the deviation between the actual track and the preset track , which is the Euclidean distance formula, used to calculate the straight-line distance between two points in three-dimensional space, where e: the Euclidean distance between the two points (the final calculation result, representing the straight-line length between the two points in space); : The spatial coordinates of the target point (here is the current position of the drone); : is the spatial coordinate of the reference point (or desired point) (here is the target waypoint of the UAV); the PID control algorithm is used, and the control law is ,in : is the controller output (the final calculation result, i.e. the motor speed of the drone), : is the proportional coefficient (adjusts the response strength of the "current error", the larger it is, the more sensitive it is to the error), : is the error at the current moment (the deviation between the actual position of the drone and the target position), : is the integral coefficient (adjusts the correction strength of "historical error accumulation" to eliminate long-term static errors), : is the differential coefficient (adjusts the response intensity of the "error change trend" and is used to predict the error), : is the integral term of the error (error About time The integral of , representing the “cumulative sum of historical errors”), : is the differential term of the error (error About time The first derivative of represents the "rate of error change." When encountering sustained strong crosswinds, the anti-interference navigation mode is activated, adding the anti-interference compensation term Uanti-wind to the control law to maintain flight stability. When the emergency landing protocol is activated, an alternate landing path is planned based on the surrounding geographic environment (e.g., terrain and buildings obtained through the Geographic Information System (GIS)) and the remaining battery power (Eremain), selecting the nearest and safest alternate landing location.

[0108] Furthermore, the specific implementation of this embodiment includes:

[0109] Control Algorithm Implementation: In the flight control module, a precise speed-pitch coordinated control algorithm is developed based on the quadrotor coupled dynamics model. This algorithm precisely adjusts motor speed and pitch angle based on the preset route and real-time weather data, achieving high-precision track tracking.

[0110] The speed-pitch angle coordinated control algorithm and the anti-interference navigation mode algorithm are written in C on the flight control module's hardware platform (e.g., a flight control board based on the STM32H7 series microcontroller). The control algorithm's sampling frequency is set to 100Hz to track the drone's flight status in real time. A hardware timer generates a precise PWM signal with 12-bit resolution to control the motor speed. The drone's attitude angle is measured in real time using an inertial measurement unit (IMU). The IMU uses a six-axis or nine-axis sensor, such as the MPU9250, and is connected to the flight control board via the I2C bus.

[0111] Emergency Mode Trigger: When sensors detect sustained strong crosswinds and abnormal flight attitude, the flight control module automatically activates anti-interference navigation mode. In this mode, the control algorithm strengthens adjustments to motor speed and pitch angle to maintain stable flight. Simultaneously, the emergency landing protocol is triggered, planning an alternate landing path based on the surrounding geographical environment and the drone's status, and alerting the operator.

[0112] The sensor detects a continuous strong crosswind when the crosswind speed Vside exceeds the set threshold (e.g. 10m / s) and the duration exceeds the critical value t (e.g. 5 seconds). When the conditions are met, the flight control module activates the anti-interference navigation mode within 100 milliseconds and sends an alert message to the operator through the wireless communication module (e.g. 4G module), including the drone's location, remaining battery power, current weather conditions, etc. The alternate landing path planning algorithm uses The algorithm, combined with geographic information system data, plans the optimal alternate landing path within 5 seconds.

[0113] The mission effectiveness compensation unit 105 is used for mission effectiveness compensation: after completing the avoidance action, the flight path is replanned based on the improved genetic algorithm and combined with the time and space constraints; the flight speed is reasonably adjusted through the speed increment allocation mechanism to eliminate the delay caused by the avoidance and ensure on-time delivery. Figure 4 , specifically:

[0114] The initial population consists of randomly generated logistics path chromosomes. Each chromosome encodes a possible logistics path, and the path encoding method uses a node sequence representation. The fitness function f comprehensively considers factors such as path length L, delivery time T, and remaining power E. For example, ,in 、 、 The population is continuously evolved through selection, crossover, and mutation operations. The selection operation uses the roulette wheel selection method, the crossover operation uses the partial matching crossover (PMX) or sequential crossover (OX) method, and the mutation operation uses the exchange mutation or insertion mutation method. After multiple iterations, the optimal logistics path that meets the time and space constraints is found. The speed increment allocation mechanism calculates the speed increment of each section based on the newly planned path and the remaining delivery time Tremain. , the formula is ,in is the road distance, Adjust the drone's flight speed within the originally planned speed and safety limits to eliminate delivery delays.

[0115] Furthermore, the specific implementation of this embodiment includes:

[0116] The mission efficiency compensation unit 105 performs route replanning. After the avoidance maneuver is completed, the mission efficiency compensation module is activated. This module uses a modified genetic algorithm to replan the logistics route based on spatiotemporal constraints such as the drone's current location, remaining battery power, destination, and time limit. Through multiple iterative calculations, the optimal route solution is found.

[0117] The mission effectiveness compensation module is based on an improved genetic algorithm, with an initial population size of 50 and an evolutionary number of 100. Geographic Information System (GIS) data is used to obtain information about the drone's current location, destination, and surrounding geographic environment, such as obstacle locations and no-fly zones. This information is incorporated into the fitness function as a constraint to ensure the planned path is safe and feasible.

[0118] The mission efficiency compensation module 105 performs speed optimization: Based on the newly planned route and remaining delivery time, the mission efficiency compensation module uses a speed increment allocation mechanism to rationally adjust the drone's flight speed along different sections of the route. Where safety permits, the flight speed is appropriately increased to compensate for delivery delays and ensure that logistics tasks are completed on time.

[0119] The speed increment allocation mechanism is activated after the path replanning is completed. The speed increment of each section is calculated based on the new planned path and the remaining delivery time. In practical applications, considering the power performance and flight safety of the UAV, the speed increment upper limit is set to By adjusting the drone's throttle control signal, speed optimization is achieved, minimizing delivery delays while ensuring flight safety.

[0120] This invention can more reliably enhance drone logistics service capabilities in complex weather conditions, demonstrating significant practical value and innovation. It is not only applicable to drone logistics but also to other drone application scenarios, such as agriculture and surveillance, demonstrating its broad potential for application.

[0121] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above-mentioned preferred embodiments are only used to illustrate the present invention and are not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A UAV control method considering weather risks, characterized in that: The method comprises the following steps: a. Real-time UAV flight environment data is collected through an onboard three-axis anemometer and temperature and humidity sensor. Cloud evolution maps from meteorological satellites and 50-meter resolution wind shear warning signals from ground-based radar are simultaneously received, and multimodal meteorological data is processed. The UAV flight environment data includes wind speed and direction, temperature, and humidity data within the UAV flight environment. b. Use a spatiotemporal convolutional neural network to fuse and analyze multimodal meteorological data, predict the storm intensity distribution within the track area in the next 2-5 minutes, and generate a three-dimensional threat level grid map; c. For conventional meteorological risks, a threat level-avoidance strategy mapping table is constructed. The threat level is determined based on the generated three-dimensional threat level grid map, and the corresponding avoidance strategy is called from the threat level-avoidance strategy mapping table to guide the UAV to perform the corresponding avoidance action; d. In extreme situations, the aircraft automatically activates the anti-interference navigation mode, which uses a quadrotor coupled dynamics model to enhance flight stability through coordinated speed-pitch angle control and trigger an emergency landing protocol. e. After completing the avoidance maneuvers in steps c and / or d, replan the flight path based on the improved genetic algorithm, combined with spatiotemporal constraints, and rationally adjust the flight speed through the speed increment allocation mechanism to eliminate delays caused by the avoidance.

2. The method according to claim 1, wherein The step a specifically includes: Step S11, sensor installation and configuration: install the three-axis anemometer and temperature and humidity sensor at appropriate locations on the drone body to ensure that they can accurately sense external meteorological parameters; Step S12, data transmission and processing: The data collected by the sensor and the data received by the communication module are transmitted in real time to the system data processing background via the wireless transmission module; the system data processing background performs preliminary screening and preprocessing on the data to remove abnormal data, providing a reliable data foundation for subsequent threat modeling; Step S13, regular calibration and maintenance: send the sensor to a professional calibration agency for calibration regularly to ensure measurement accuracy; at the same time, check the signal strength and data transmission stability of the communication module, and replace aging or faulty components in a timely manner.

3. The method according to claim 1, wherein The step b specifically includes: Step S21, model training: Before the UAV is put into use, a large amount of meteorological data from different regions and weather conditions is collected, including historical meteorological data and simulated data; the spatiotemporal convolutional neural network is trained using this meteorological data, and the model parameters are adjusted to enable it to accurately predict the distribution of storm intensity; Step S22, perform real-time prediction: When the drone is flying, the background processing server packages the real-time received meteorological monitoring data into a data tensor that conforms to the model input format and transmits it to the trained spatiotemporal convolutional neural network model in real time. The spatiotemporal convolutional neural network model outputs a three-dimensional threat level grid map and returns it to the hierarchical decision module.

4. The method according to claim 3, wherein The threat level-avoidance strategy mapping table includes: When the threat level is 1, dynamic heading compensation of ±5° is implemented to fine-tune the flight direction to cope with minor weather interference; when the threat level is 2, the altitude switching protocol is activated to avoid the dangerous area by changing the flight altitude; when the threat level is 3, a B-spline flight path with energy constraints is generated, thereby ensuring that the drone can safely avoid severe weather while reasonably controlling energy consumption.

5. The method according to claim 4, wherein The step c specifically includes: Step S31, establishing a mapping table: In the decision module, based on a large amount of experimental and actual flight data, a detailed threat level-avoidance strategy mapping table is established to clearly define specific operational instructions corresponding to different threat levels; the operational instructions include: heading adjustment angle, altitude layer switching range, and detour path generation rules; Step S32, decision execution: After receiving the three-dimensional threat level grid map, the decision module quickly determines the threat level and calls the corresponding avoidance strategy from the mapping table. The strategy is transmitted to the flight control module through control instructions to guide the UAV to perform the corresponding avoidance action.

6. The method according to claim 5, wherein The step d specifically includes: Step S41, control algorithm implementation: In the flight control module, a precise speed-pitch angle coordinated control algorithm is written based on the quadrotor coupling dynamics model. This algorithm accurately adjusts the motor speed and pitch angle according to the preset route and real-time weather data to achieve high-precision track tracking. Step S42, emergency mode triggering: When the sensor detects a continuous strong crosswind and the flight attitude is abnormal, the flight control module automatically activates the anti-interference navigation mode; in this mode, the control algorithm strengthens the adjustment of the motor speed and pitch angle to keep the drone flight stable; at the same time, the emergency landing protocol is triggered, and the alternate landing path is planned according to the surrounding geographical environment and the status of the drone, and an alarm is sent to the operator.

7. The method according to claim 6, wherein The step e specifically includes: Step S51, path replanning: After the avoidance action is completed, the mission efficiency compensation module is activated. The mission efficiency compensation module uses an improved genetic algorithm to replan the logistics path according to the spatiotemporal constraints to obtain the optimal path solution. The spatiotemporal constraints include the current location of the UAV, the remaining battery power, the destination, and the time limit. Step S52, speed optimization: According to the newly planned route and the remaining delivery time, the flight speed of the drone in different sections is reasonably adjusted. If safety permits, the flight speed is appropriately increased to compensate for delivery delays and ensure that the logistics task is completed on time.

8. A UAV control system that takes weather risks into consideration, characterized in that: The system includes a meteorological data processing unit, a threat level map construction unit, an avoidance strategy decision unit, an extreme situation processing unit, and a mission effectiveness compensation unit, among which: The meteorological data processing unit is used to collect real-time UAV flight environment data through an onboard three-axis anemometer and temperature and humidity sensors, and simultaneously receive cloud evolution maps from meteorological satellites and 50-meter resolution wind shear warning signals from ground radars to process and obtain multimodal meteorological data; the UAV flight environment data includes: wind speed and direction, temperature and humidity data in the UAV flight environment; The threat level map construction unit is used to use a spatiotemporal convolutional neural network to perform fusion analysis on multimodal meteorological data, predict the storm intensity distribution within the track area in the next 2-5 minutes, and generate a three-dimensional threat level grid map; The avoidance strategy decision unit is used to construct a threat level-avoidance strategy mapping table for conventional meteorological risks, determine the threat level based on the generated three-dimensional threat level grid map, and call the corresponding avoidance strategy from the threat level-avoidance strategy mapping table to guide the UAV to perform the corresponding avoidance action; The extreme situation processing unit is used to automatically activate the anti-interference navigation mode when encountering an extreme situation, enhance flight stability through speed-pitch angle coordinated control based on the quadrotor coupling dynamics model, and trigger the emergency landing protocol; The mission effectiveness compensation unit is used to re-plan the flight path after completing the avoidance action based on the improved genetic algorithm and combined with time and space constraints, and to reasonably adjust the flight speed through the speed increment distribution mechanism to eliminate the delay caused by the avoidance.

Images