Control system for combined flight of multiple unmanned aerial vehicles for coping with wind power change

Abstract

The invention discloses a multi-unmanned aerial vehicle combined flight control system coping with wind power change, and relates to unmanned aerial vehicle flight control, and the system comprises a flight data collection module which is used for collecting unmanned aerial vehicle flight data in real time, and the unmanned aerial vehicle flight data comprises wind power data, attitude angle data and height data of each unmanned aerial vehicle; the data receiving and processing module is electrically connected with the flight data acquisition module, and the data receiving and processing module is used for receiving unmanned aerial vehicle flight data acquired in real time and performing noise removal on the acquired data. According to the multi-unmanned aerial vehicle combined flight control system provided by the invention, the wind field prediction model is constructed by combining the LSTM neural network with the time attention mechanism, wind power periodic change characteristics are accurately captured, real-time correction is performed in cooperation with the Kalman filter, the prediction precision and timeliness are both optimized, and the system is suitable for large-scale popularization and application. A multi-unmanned aerial vehicle cooperative kinetic model is constructed based on a prediction result, so that the formation trajectory tracking error is reduced.

Description

Technical Field

[0001] The present invention relates to unmanned aerial vehicle (UAV) flight control, and in particular to a control system for combined flight of multiple UAVs in response to wind changes. Background Art

[0002] With the widespread application of drone formation technology in disaster relief, logistics and transportation, environmental monitoring and other fields, multi-machine collaborative flight systems face severe challenges in complex atmospheric environments. Wind changes are a core factor affecting the stability of drone formations. Their randomness and time-varying nature place multi-dimensional demands on flight control, involving multiple technical links such as wind field perception, dynamic modeling, and collaborative decision-making. The current mainstream technology achieves basic attitude control through the fusion of inertial navigation and satellite positioning, and combines aerodynamic parameter correction strategies for wind resistance compensation. However, in scenarios with sudden changes in the three-dimensional wind field, existing systems still face key bottlenecks in prediction accuracy and response speed, making it difficult to meet the stringent requirements of high-density formation flight for real-time collaborative control.

[0003] In the existing technology, multi-UAV systems often use a fixed-gain PID controller combined with a feedforward wind disturbance compensation method, but phase lag and overshoot are prone to occur in non-steady-state wind fields. Traditional wind field prediction models mostly rely on single-point wind speed time series analysis and lack effective modeling of spatial vortex structure and shear gradient, resulting in a spatiotemporal mismatch between control command generation and the actual wind field evolution. At the same time, existing formation maintenance algorithms often use rigid constraint strategies under strong wind interference, which limits the autonomous wind avoidance maneuverability of UAVs and easily causes chain formation disorder. At the communication and positioning level, conventional GPS and inertial navigation fusion solutions are susceptible to multipath effects in complex electromagnetic environments, resulting in the accumulation of errors in the relative position estimation of the formation. To this end, we propose a control system for the combined flight of multiple UAVs to cope with wind changes. Summary of the Invention

[0004] In order to solve the above technical problems, a control system for the combined flight of multiple UAVs that can cope with wind changes is provided. This technical solution solves the above problems.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is: A control system for combined flight of multiple UAVs in response to wind changes, comprising: The flight data acquisition module is used to collect UAV flight data in real time, and the UAV flight data includes: wind data, attitude angle data and altitude data of each UAV; The data receiving and processing module is electrically connected to the flight data acquisition module, and is used to receive the real-time collected UAV flight data and remove noise from the collected data; The model building and instruction generation module is electrically connected to the data receiving and processing module, and is used to build a wind field prediction model based on the processed UAV flight data to predict future wind speed and wind direction, and generate wind resistance control instructions based on the wind field prediction results; The UAV control and flight module is electrically connected to the model building and instruction generation module, and is used to send the generated control instructions to the UAV flight controller to control the attitude and flight trajectory of the UAV.

[0006] Preferably, the wind data collected by the flight data collection module include: wind size and direction, and the attitude angle data collected include: roll angle, pitch angle and yaw angle, and the flight altitude of the drone is obtained in real time through the altitude sensor.

[0007] Preferably, the method for the data receiving and processing module to remove noise from the collected data is: For wind data, the original signal is decomposed by multi-scale wavelet to obtain high-frequency detail coefficients and low-frequency approximate coefficients. The high-frequency coefficients are soft-thresholded and then the denoised signal is reconstructed by inverse wavelet transform. The Kalman filter is used for attitude angle data and altitude data as follows: based on the current attitude angle and altitude state of the UAV, combined with its dynamic model and control input, the attitude angle and altitude value at the next moment are predicted to obtain the predicted value, the current attitude angle and altitude state are compared with the predicted value, and the difference between the two is calculated. The filter weight is dynamically adjusted according to the sensor measurement accuracy and the credibility of the model prediction, and the predicted value and the measured value are weightedly fused using the weight to obtain the optimal estimation result.

[0008] Preferably, the dynamic weight adjustment method of the Kalman filter is specifically as follows: A time-varying model of sensor measurement error is established, and the variance matrix of attitude sensor output is calculated in real time using a sliding window statistical method. A model credibility evaluation function based on the Lyapunov exponent is constructed, and the process noise covariance matrix Q and the observation noise covariance matrix R are dynamically adjusted according to flight state parameters. A variable gain adjustment strategy is designed to increase the model prediction weight when the UAV is in a high-maneuverability state and to increase the sensor measurement weight during steady-state flight. The Kalman gain matrix is updated in real time by solving the Riccati differential equation, and the filter parameters are adaptively adjusted.

[0009] Preferably, the specific method for constructing the wind field prediction model in the model construction and instruction generation module is: A spatiotemporal sequence database is established based on processed historical wind data, using a long short-term memory neural network as the basic prediction framework. The input layer uses a dual-channel input system to receive wind speed scalar data and wind direction vector data. The hidden layer incorporates a temporal attention mechanism to capture the periodic characteristics of wind changes. The output layer generates a three-dimensional spatial wind field distribution matrix for the future through a fully connected network. A real-time correction unit based on the Kalman filter is established to dynamically compensate the errors between the measured wind field data at the current moment and the predicted results, and generate a wind field evolution map with confidence intervals. The map contains: wind speed gradient, vortex intensity and change trend parameters of each spatial grid point.

[0010] Preferably, the method for generating the wind resistance control instruction specifically includes: Based on the three-dimensional wind field distribution matrix output by the wind field prediction model, a multi-UAV collaborative dynamics model is established. The torque equation between the wind pressure center and the center of gravity is calculated based on the aerodynamic parameters, mass distribution and current flight status of each UAV. A rolling optimization framework based on model predictive control (MPC) is constructed to solve a multi-objective optimization function including attitude stability, trajectory tracking accuracy, and energy consumption at a fixed update frequency. The rudder deflection, rotor speed adjustment, and thrust vector distribution scheme for each UAV are generated, and collision avoidance constraints between UAVs are established. The optimal control sequence for each UAV is calculated through a distributed optimization algorithm, and a phase-synchronized instruction set with a timestamp is generated.

[0011] Preferably, the specific method for the UAV control and flight module to achieve multi-machine collaborative control is: establishing a formation maintenance algorithm based on relative kinematics, and calculating the relative position constraint equations between each machine in real time according to the preset formation topology structure; designing a double closed-loop control structure, the inner loop adopts a quaternion attitude controller to achieve rapid response in the body coordinate system, and the outer loop adopts a nonlinear trajectory tracking controller to handle the track correction in the inertial coordinate system; starting the formation reconstruction mechanism under strong wind disturbance conditions, adjusting the formation density parameters through a dynamic weight distribution algorithm, and automatically switching to the elastic formation mode when it is predicted that there will be discontinuous shear exceeding the threshold in the wind field and reserving space for each machine to autonomously avoid the wind, while maintaining the connectivity constraints of the overall formation.

[0012] Preferably, the system also includes a distributed communication coordination module, which uses the time division multiple access (TDMA) protocol to establish communication links between drone groups and designs a packet transmission mechanism including data priority marking, in which attitude control instructions have the highest transmission priority, and wind field prediction data uses forward error correction coding; A relative positioning network based on UWB technology is constructed. Each drone is equipped with a dual-band transceiver to achieve centimeter-level relative position measurement. The GPS positioning data and UWB ranging information are integrated through an extended Kalman filter to establish an anti-occlusion hybrid positioning system.

[0013] Preferably, the system sets up multiple redundant control mechanisms, including: In the data receiving and processing module, two pre-processing channels, wavelet threshold denoising and empirical mode decomposition (EMD), are run in parallel, and the optimal denoising result is selected using a Mahalanobis distance detector. In the model building module, a wind farm simulator based on a physical model and a data-driven prediction model are simultaneously maintained. When the difference between the two outputs exceeds a threshold, an expert system is activated for arbitration. A three-level fault-tolerant architecture is set up at the control execution layer. The main controller uses FPGA to implement hard real-time control, the backup controller runs a simplified control algorithm based on a microkernel RTOS, and the emergency controller directly executes the preset stable return strategy. Switching between each layer is achieved through heartbeat detection.

[0014] Preferably, the system also includes an abnormal condition processing module, which monitors the health status indicators of each subsystem in real time and initiates a hierarchical response strategy when a persistent abnormality is detected: For instantaneous wind exceeding the limit, a gust through control sequence is generated to reduce the frontal area through a preset pitch-roll coupled maneuvering mode; When a continuous strong wind threat is predicted, the formation's overall altitude migration strategy is activated, and the best shelter altitude layer is selected based on the terrain database; Under extreme wind shear conditions, drones in the control center release tracer particles to build a local wind field observation network, while peripheral drones automatically deploy aerodynamic deceleration surfaces to form a windbreak. When a multi-machine system faces the risk of disintegration, it is dynamically reconstructed into multiple sub-formations according to preset priorities. Each sub-formation independently executes the wind avoidance strategy and maintains the minimum safety unit.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The control system for the combined flight of multiple UAVs proposed in the present invention constructs a wind field prediction model through the LSTM neural network combined with the temporal attention mechanism, accurately captures the periodic change characteristics of wind force, and cooperates with the Kalman filter for real-time correction, so that the prediction accuracy and timeliness are optimized. The multi-UAV collaborative dynamics model constructed based on the prediction results reduces the formation trajectory tracking error and improves the attitude stability accuracy. At the same time, it optimizes energy consumption distribution to extend the flight life. The triple redundant control architecture covers data preprocessing, model construction and execution layers, and combines the graded abnormal working condition response strategy to effectively respond to emergencies such as instantaneous wind force exceeding the limit. The UWB and GPS fusion positioning technology is used to achieve centimeter-level relative position measurement, which greatly improves the reliability of formation flight. The control system shows excellent adaptability, stability and safety under complex wind field conditions, providing a strong technical guarantee for the collaborative operation of multiple UAVs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a system module framework diagram of the present invention; Figure 2 It is a system workflow diagram of the present invention. DETAILED DESCRIPTION

[0017] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0018] Reference Figure 1 As shown, a control system for the combined flight of multiple UAVs in response to wind changes includes: a flight data acquisition module, a data receiving and processing module, a model building and command generation module, and a UAV control and flight module; its specific implementation steps are as follows: The system's core architecture includes a flight data acquisition module, a data reception and processing module, a model building and command generation module, a UAV control and flight module, and a distributed communication coordination module and redundant control mechanism. The flight data acquisition module collects real-time flight data from each UAV using a six-axis accelerometer, gyroscope, and laser altimeter. Wind data, including wind speed magnitude and direction vector, is acquired via a pitot tube anemometer. Attitude angle data includes roll angle φ, pitch angle θ, and yaw angle ψ. Altitude data is output by the laser altimeter. The data acquisition frequency is set to 200 Hz to ensure that high-frequency dynamic characteristics can be captured even in strong wind disturbances.

[0019] The data receiving and processing module uses a parallel dual-channel preprocessing mechanism. For wind data, the db4 wavelet basis is used for 5-layer decomposition, and a soft threshold function is applied to the high-frequency coefficients: in, is the soft threshold function, is the high frequency coefficient to be processed, is the dynamic threshold, threshold , is the noise standard deviation, is the signal length. In this embodiment, The first-order intrinsic mode function (IMF) is calculated by empirical mode decomposition (EMD), and N is 1024 points. The attitude angle and height data are obtained by using the improved Kalman filter algorithm, and its state equation is: Where, is the state vector, The discrete-time system A moment, is the state transition matrix, is the control input matrix, is the control input, is the process noise, which includes 12-dimensional state vectors such as position, velocity, and attitude angle. When dynamically adjusting the process noise covariance matrix Q and the observation noise covariance matrix R, the Lyapunov index is used. Calculate model credibility: Where, is the Lyapunov exponent, is the number of time steps, is the dynamic function of the system, is the state vector A quantity, is the partial derivative of the dynamic function with respect to the state component, combined with the attitude sensor variance σ² of the sliding window statistics, through the adaptive weight and Adjust the gains of Q and R. T is the sliding window length, which is set to 0.5 seconds, corresponding to 100 sampling points.

[0020] The model building and instruction generation module builds a wind field prediction model based on the LSTM neural network, and the input layer receives the standardized wind speed sequence. and wind direction sequence The hidden layer uses a bidirectional LSTM structure to capture spatiotemporal features. The output layer generates a three-dimensional wind field matrix for the next five seconds through a fully connected layer, representing the wind speed vectors at 100 spatial grid points within five time steps. The network is trained using the Adam optimizer with a learning rate of 0.001, a batch size of 64, and 1000 training cycles. The real-time correction unit uses a Kalman filter to correct the prediction results: Where, is the corrected weight, is the prediction weight, K is the Kalman gain matrix, is the measurement weight, and H is the observation matrix. In this embodiment, K is updated in real time by solving the Riccati equation to ensure that the prediction error covariance matrix is minimized.

[0021] When generating wind-resistant control instructions, based on the multi-UAV collaborative dynamics model: Where, is the mass matrix, which is used to describe the inertial characteristics of the system and is related to the mass and moment of inertia of the drone. is the generalized acceleration vector, is the inertial force, are the Coriolis force and centripetal force matrices, is the gravity term, is the control input vector, is the wind load disturbance term; An MPC rolling optimization framework is established to solve the objective function including attitude stabilization, trajectory tracking, and energy consumption: Where, is the objective function, To predict the time domain length, For the Step tracking error, For the Step control input, is the terminal error, 、 、 is the weight matrix, and N = 10 is the prediction time domain. In this embodiment, the diagonal elements of Q are 100 (position), 10 (velocity), and 50 (attitude angle), R is set to the identity matrix of 0.1, and P is 5 times Q. The optimization problem is solved in a distributed manner using the alternating direction multiplier method (ADMM), with control commands updated every 10 ms.

[0022] The drone control and flight module adopts a dual closed-loop control structure, and the inner loop attitude control is based on quaternion error: Where, To control the torque, is the proportional gain, is the quaternion attitude error, is the differential gain, is the angular velocity error, , , the quaternion error is obtained by rotating the vector. The outer loop trajectory tracking is designed using the backstepping method: Where, is the expected acceleration, is the target position, is the actual location, and Set to 3 and 1.5 respectively. The formation keeping algorithm uses the relative kinematic equation: Real-time calculation of each machine's position constraints, including: is the follower target position, is the current position of the navigator, For drones arrive The rotation matrix of is the preset relative displacement. In this embodiment, the formation adopts a diamond structure, and d_ji is set to 1.5 times the wingspan length based on the aerodynamic characteristics of the aircraft model. When the predicted wind shear exceeds 15m / s, the elastic formation mode is activated. By adjusting the formation density parameter γ∈[0.5,1.5], each aircraft is allowed to autonomously avoid obstacles within a range of ±0.5m. At the same time, the virtual spring damping model is used to maintain the connectivity of the formation. Its expression is: Where, For virtual combined force, and is the drone position vector, is the expected formation distance, and is the velocity vector of the drone.

[0023] The distributed communication coordination module uses the TDMA protocol for time slot allocation, with a total bandwidth of 50Mbps and a time slot length of 2ms, supporting up to 128 nodes. Attitude control commands are assigned a priority of 0x01, and preemptive scheduling is used to ensure transmission within 1ms. Wind farm forecast data uses (255,223) RS encoding, which can correct 16 symbol errors. The relative positioning system uses the UWB ranging equation: Where, is the distance between nodes, is the speed of light, The signal propagation time difference is combined with the extended Kalman filter to fuse GPS and UWB data, and the positioning accuracy reaches ±0.1m. In this embodiment, the UWB module operates in the 3.1-10.6GHz frequency band and adopts the DS-UWB modulation method. The ranging error .

[0024] In the multiple redundancy mechanism, the data preprocessing module runs wavelet denoising and EMD decomposition in parallel, using the Mahalanobis distance: Where, is the Mahalanobis distance, is the multidimensional variable to be evaluated, is the mean vector, is the matrix and vector transpose operator, is the inverse of the covariance matrix.

[0025] The optimal denoising result is selected. The Mahalanobis distance threshold is set to 3; exceeding this value triggers expert system arbitration. The control execution layer employs a three-level fault-tolerant architecture. The main controller uses a Xilinx Zynq FPGA for 1kHz hard real-time control. The backup controller runs the μC / OS-II system with a 10ms execution cycle. The emergency controller executes a pre-set three-stage return-to-home strategy: first, lowering the flight altitude to 100m, then adjusting the heading to the headwind, and finally activating the parachute system.

[0026] The abnormal operating condition processing module monitors the status of each subsystem in real time. When it detects a wind speed exceeding 20m / s for three seconds, it initiates the formation altitude migration strategy, selecting a sheltered altitude layer at an altitude of 300-500m using the terrain database. The terrain database is built based on DEM data with a resolution of 5m and includes parameters such as slope and roughness. During the altitude migration process, a variable gain PID controller is used: Where, To control the command output, is the integral gain, is the height error, =0.5, =0.1, =0.05, eliminating altitude errors through feedforward compensation. Under extreme wind shear conditions, the control center's drones released 300 tracer particles to establish a local wind field observation network. The tracer particle density was set to 0.1 particles / m³, and wind field data was acquired through real-time lidar scanning. Peripheral drones deployed a 0.5m² aerodynamic deceleration surface, forming a windbreak with a radius of 10m. The deceleration surface used a NACA0012 airfoil with a maximum lift-to-drag ratio of 12. When the system faced the risk of disintegration, the 10-aircraft formation was reconstructed into three independent sub-formations according to preset priorities. Each sub-formation maintained a minimum safety unit of at least three aircraft, and autonomously selected a wind avoidance strategy using a distributed optimization algorithm. Prioritization was based on remaining battery capacity, payload importance, and communication quality, and was dynamically evaluated using a fuzzy logic system.

[0027] Through the above-mentioned technical means, this embodiment achieves a formation flight error of less than 0.3m, an attitude stabilization accuracy of 0.5°, and a system power consumption reduction of 15% in an environment with an average wind speed of 12m / s and a gust of 18m / s, effectively improving the collaborative flight capability of the multi-UAV system in complex wind fields.

[0028] The present invention uses onboard sensors to collect wind speed, attitude, and altitude data in real time. After wavelet and Kalman filtering, the data is fed into an LSTM neural network to predict future wind field distribution. The control system constructs a collaborative dynamics model based on the prediction results, generates wind resistance commands using an MPC algorithm, and adjusts the rotor speed and thrust vector of each aircraft. During formation flight, dual closed-loop control is used to maintain formation, and formation density is dynamically adjusted in the event of wind shear. The system achieves centimeter-level positioning through UWB communication, combined with redundant fault-tolerant mechanisms to ensure flight safety. In extreme wind conditions, obstacle avoidance strategies are activated or sub-formations are reconfigured.

[0029] In summary, the advantages of the present invention are: real-time wind field prediction and dynamic wind resistance control improve formation stability, multi-UAV collaborative dynamics model realizes precise trajectory tracking, flexible formation strategy adapts to complex wind conditions, UWB centimeter-level positioning ensures formation accuracy, multiple redundancy mechanisms enhance system reliability, energy consumption optimization algorithm extends flight time, and graded obstacle avoidance strategy in extreme wind conditions ensures flight safety.

[0030] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A control system for combined flight of multiple UAVs in response to wind changes, characterized in that: include: The flight data acquisition module is used to collect UAV flight data in real time, and the UAV flight data includes: wind data, attitude angle data and altitude data of each UAV; The data receiving and processing module is electrically connected to the flight data acquisition module, and is used to receive the real-time collected UAV flight data and remove noise from the collected data; The model building and instruction generation module is electrically connected to the data receiving and processing module, and is used to build a wind field prediction model based on the processed UAV flight data to predict future wind speed and wind direction, and generate wind resistance control instructions based on the wind field prediction results; The UAV control and flight module is electrically connected to the model building and instruction generation module, and is used to send the generated control instructions to the UAV flight controller to control the attitude and flight trajectory of the UAV.

2. A control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The wind data collected by the flight data collection module include: wind size and direction, and the attitude angle data collected include: roll angle, pitch angle and yaw angle. The flight altitude of the UAV is obtained in real time through the altitude sensor.

3. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The method for the data receiving and processing module to remove noise from the collected data is: For wind data, the original signal is decomposed by multi-scale wavelet to obtain high-frequency detail coefficients and low-frequency approximate coefficients. The high-frequency coefficients are soft-thresholded and then the denoised signal is reconstructed by inverse wavelet transform. The Kalman filter is used for attitude angle data and altitude data as follows: based on the current attitude angle and altitude state of the UAV, combined with its dynamic model and control input, the attitude angle and altitude value at the next moment are predicted to obtain the predicted value, the current attitude angle and altitude state are compared with the predicted value, and the difference between the two is calculated. The filter weight is dynamically adjusted according to the sensor measurement accuracy and the credibility of the model prediction, and the predicted value and the measured value are weightedly fused using the weight to obtain the optimal estimation result.

4. A control system for combined flight of multiple UAVs in response to wind changes according to claim 3, characterized in that: The dynamic weight adjustment method of the Kalman filter is specifically as follows: A time-varying model of sensor measurement error is established, and the variance matrix of attitude sensor output is calculated in real time using a sliding window statistical method. A model credibility evaluation function based on the Lyapunov exponent is constructed, and the process noise covariance matrix Q and the observation noise covariance matrix R are dynamically adjusted according to flight state parameters. A variable gain adjustment strategy is designed to increase the model prediction weight when the UAV is in a high-maneuverability state and to increase the sensor measurement weight during steady-state flight. The Kalman gain matrix is updated in real time by solving the Riccati differential equation, and the filter parameters are adaptively adjusted.

5. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The specific method for constructing the wind field prediction model in the model construction and instruction generation module is: A spatiotemporal sequence database is established based on processed historical wind data, using a long short-term memory neural network as the basic prediction framework. The input layer uses a dual-channel input system to receive wind speed scalar data and wind direction vector data. The hidden layer incorporates a temporal attention mechanism to capture the periodic characteristics of wind changes. The output layer generates a three-dimensional spatial wind field distribution matrix for the future through a fully connected network. A real-time correction unit based on the Kalman filter is established to dynamically compensate the errors between the measured wind field data at the current moment and the predicted results, and generate a wind field evolution map with confidence intervals. The map contains: wind speed gradient, vortex intensity and change trend parameters of each spatial grid point.

6. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The method for generating the wind resistance control instruction specifically includes: Based on the three-dimensional wind field distribution matrix output by the wind field prediction model, a multi-UAV collaborative dynamics model is established. The torque equation between the wind pressure center and the center of gravity is calculated based on the aerodynamic parameters, mass distribution and current flight status of each UAV. A rolling optimization framework based on model predictive control (MPC) is constructed to solve a multi-objective optimization function including attitude stability, trajectory tracking accuracy, and energy consumption at a fixed update frequency. The rudder deflection, rotor speed adjustment, and thrust vector distribution scheme for each UAV are generated, and collision avoidance constraints between UAVs are established. The optimal control sequence for each UAV is calculated through a distributed optimization algorithm, and a phase-synchronized instruction set with a timestamp is generated.

7. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The specific method for the UAV control and flight module to achieve multi-machine collaborative control is as follows: establishing a formation maintenance algorithm based on relative kinematics, and calculating the relative position constraint equations between each machine in real time according to the preset formation topology structure; designing a dual closed-loop control structure, the inner loop adopts a quaternion attitude controller to achieve rapid response in the body coordinate system, and the outer loop adopts a nonlinear trajectory tracking controller to handle track correction in the inertial coordinate system; activating the formation reconstruction mechanism under strong wind disturbance conditions, adjusting the formation density parameters through a dynamic weight distribution algorithm, and automatically switching to the elastic formation mode when the wind field is predicted to have discontinuous shear exceeding the threshold and reserving space for each machine to autonomously avoid the wind while maintaining the connectivity constraints of the overall formation.

8. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The system also includes a distributed communication coordination module, which uses the Time Division Multiple Access (TDMA) protocol to establish communication links between drone groups. It also designs a packet transmission mechanism that includes data priority marking, where attitude control commands have the highest transmission priority, and wind field prediction data uses forward error correction coding. A relative positioning network based on UWB technology is constructed. Each drone is equipped with a dual-band transceiver to achieve centimeter-level relative position measurement. The GPS positioning data and UWB ranging information are integrated through an extended Kalman filter to establish an anti-occlusion hybrid positioning system.

9. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The system is equipped with multiple redundant control mechanisms, including: In the data receiving and processing module, two pre-processing channels, wavelet threshold denoising and empirical mode decomposition (EMD), are run in parallel, and the optimal denoising result is selected using a Mahalanobis distance detector. In the model building module, a wind farm simulator based on a physical model and a data-driven prediction model are simultaneously maintained. When the difference between the two outputs exceeds a threshold, an expert system is activated for arbitration. A three-level fault-tolerant architecture is set up at the control execution layer. The main controller uses FPGA to implement hard real-time control, the backup controller runs a simplified control algorithm based on a microkernel RTOS, and the emergency controller directly executes the preset stable return strategy. Switching between each layer is achieved through heartbeat detection.

10. The control system for combined flight of multiple UAVs in response to wind changes according to claim 1, characterized in that: The system also includes an abnormal condition processing module, which monitors the health status indicators of each subsystem in real time and initiates a graded response strategy when persistent abnormalities are detected: For instantaneous wind exceeding the limit, a gust through control sequence is generated to reduce the frontal area through a preset pitch-roll coupled maneuvering mode; When a continuous strong wind threat is predicted, the formation's overall altitude migration strategy is activated, and the best shelter altitude layer is selected based on the terrain database; Under extreme wind shear conditions, drones in the control center release tracer particles to build a local wind field observation network, while peripheral drones automatically deploy aerodynamic deceleration surfaces to form a windbreak. When a multi-machine system faces the risk of disintegration, it is dynamically reconstructed into multiple sub-formations according to preset priorities. Each sub-formation independently executes the wind avoidance strategy and maintains the minimum safety unit.

Images