A method and system for dynamic deviation compensation of a UAV path in a strong wind environment

Abstract

This invention discloses a method and system for dynamic path deviation compensation of unmanned aerial vehicles (UAVs) in strong wind environments, relating to the field of UAV flight control technology. The method acquires real-time three-dimensional wind field information through multi-source sensor fusion processing, inputs it into a hybrid aerodynamic model composed of a physical mechanism basis model and an online neural network compensation model in parallel, and calculates the wind disturbance force and torque experienced by the UAV. Based on the wind disturbance force and torque, a feedforward compensation command is generated through a nonlinear dynamic inverse compensator considering actuator control margin constraints. A high-order sliding mode disturbance observer estimates the residual disturbance and superimposes it onto the feedforward command. After adaptive amplitude limiting, the signal is superimposed with the outputs of the position and attitude controllers to generate the final control signal. This invention improves the path tracking accuracy and flight stability of UAVs in strong wind environments, avoids actuator saturation, and is suitable for high-precision UAV flight missions in complex outdoor environments.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control technology, specifically to a method and system for dynamic path deviation compensation of UAVs in strong wind environments. Background Technology

[0002] With the rapid development of drone technology, drones have been widely used in various fields such as power line inspection, forest fire prevention, emergency rescue, and surveying and exploration. In these application scenarios, drones often need to perform long-duration, long-distance flight missions in complex outdoor environments. Strong winds are one of the most significant factors affecting drone flight safety and mission accuracy. Existing drone path deviation compensation methods in strong wind environments generally suffer from several technical defects. The wind field estimation stage often relies on a single wind speed sensor or a single filtering algorithm. Wind speed sensors are easily affected by drone fuselage obstruction, rotor downwash, and changes in the drone's own attitude, leading to significant errors in the measurement data. Single filtering algorithms have poor tracking capabilities for abrupt and turbulent wind fields, resulting in significant time delays in wind field estimation. The aerodynamic model stage often uses offline calibrated fixed physical models, which cannot adapt to aerodynamic parameter drift caused by changes in flight altitude, flight speed, and ambient temperature, nor can they accurately characterize unsteady aerodynamic effects. The existing methods suffer from several problems. First, the wind disturbance force and torque output by the model deviate significantly from the actual values. Second, the feedforward compensation stage typically calculates the compensation command directly based on the wind disturbance force without considering the residual control capability of the actuator in real time. When the wind disturbance intensity is high, the compensation command can easily exceed the maximum output range of the actuator, leading to actuator saturation, a sharp decline in control performance, and even loss of control of the UAV. Third, existing methods generally lack effective estimation and compensation mechanisms for residual disturbances that the hybrid aerodynamic model fails to characterize. These residual disturbances include unsteady wind disturbances such as gusts and turbulence, as well as aerodynamic effects not modeled in the model, which further increase the path tracking error of the UAV. These combined issues result in low path tracking accuracy and poor flight stability for UAVs in strong wind environments, failing to meet the requirements of high-precision flight missions. To address these problems, this invention proposes a method and system for dynamic path deviation compensation for UAVs in strong wind environments. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for dynamic path deviation compensation of UAVs in strong wind environments, thereby improving the path tracking accuracy and flight stability of UAVs in strong wind environments.

[0004] To achieve the above objectives, the embodiments of this invention provide the following technical solutions:

[0005] This application provides a method for dynamic path deviation compensation of a UAV in strong wind conditions, including the following steps: S1, acquiring the flight status information of the UAV and obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing; S2, inputting the real-time three-dimensional wind field information into a hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV, wherein the hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction, and the output of the wind disturbance force and torque satisfies the following formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. S3. Based on the wind disturbance force and torque, a feedforward compensation command is generated by a nonlinear dynamic inverse compensator that considers the actuator control margin constraint; S4. The residual disturbance that the hybrid aerodynamic model fails to characterize is estimated in real time using a high-order sliding mode disturbance observer, and the residual disturbance estimate is fed forward and superimposed into the feedforward compensation command to obtain a corrected feedforward compensation command; S5. The corrected feedforward compensation command is adaptively limited and then superimposed with the output commands of the original position controller and attitude controller of the UAV to generate the final control signal for driving the actuator.

[0006] Further, the step S1, obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing, specifically includes: S11, measuring local wind speed vectors based on multiple wind speed sensors mounted on the UAV arm, performing coordinate transformation in conjunction with the UAV's real-time attitude angle, and obtaining the initial wind speed vector at the UAV's center of mass through weighted least squares spatial interpolation; S12, using accelerometer and satellite positioning velocity information as observations, employing an extended Kalman filter to correct the error of the initial wind speed vector, and outputting the optimal three-dimensional wind field information. The state update equation of the extended Kalman filter is: ,in For the first Time based on the past The optimal state estimation vector obtained from the observations For the first Time based on the past The prior state estimation vector obtained from the observations For the first The Kalman gain matrix at time t. For the first The observation vector at time t, To extend the nonlinear observation function of the Kalman filter, the state estimation vector includes three-dimensional wind speed components, UAV attitude angles, and angular velocity components.

[0007] Furthermore, the neural network compensation model in S2 is an online-updated radial basis function neural network. The radial basis function neural network takes the wind speed, attitude angle, rotor speed and the residual of the current mechanism model output and the force calculated based on sensor data within the historical time window as input, and outputs the compensation increment of the aerodynamic coefficient in real time, thereby obtaining the compensation force and torque. The online update of the radial basis function neural network adopts the recursive least squares algorithm to adjust the network weights.

[0008] Furthermore, the nonlinear dynamic inverse compensator considering actuator control margin constraints in S3, when solving for feedforward compensation acceleration and angular velocity, acquires the control margin of each actuator in real time and processes it according to the following gain scheduling method: Define thrust margin: ;when At that time, the scaling factor of the vertical compensation acceleration gain is calculated according to the following formula: Otherwise, the scaling factor is 1; where This represents the actuator thrust margin, with a value ranging from 0 to 1. The maximum available thrust of a single rotor of the drone. For the first The current thrust command is constantly sent to the rotor. This is a preset critical margin threshold. This is the gain scaling factor for compensating acceleration in the vertical channel.

[0009] Furthermore, the adaptive limiting process in S5 employs a variable structure limiting filter, and the limiting threshold of the variable structure limiting filter... The fuzzy logic regulator is based on the absolute value of the rate of change of wind speed. and position tracking error Real-time determination, expressed by the formula: When wind speed changes drastically and tracking error is large, the limiting threshold is adaptively increased to allow for large instantaneous compensation; conversely, the threshold is decreased to suppress measurement noise.

[0010] Furthermore, the higher-order sliding mode perturbation observer in S4 employs the super-twisting algorithm, and the residual perturbation estimation law of the higher-order sliding mode perturbation observer is: ;in, This is the estimated vector of the residual perturbation. The first design gain for the super-twisting algorithm, For sliding mode variables absolute value Exponentiation. For symbolic functions, This is an integration operation over time. This is the second design gain for the super-twisting algorithm. The sliding mode variable is calculated from the actual acceleration signal measured by the accelerometer and the wind disturbance force output by the hybrid aerodynamic model.

[0011] Furthermore, following S5, the following step is also included: S6, when the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window, dynamic path replanning is triggered. This dynamic path replanning is based on an improved fast random expansion tree algorithm, introducing an artificial potential field bias generated by the currently estimated wind field during random sampling and node expansion. The potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

[0012] Accordingly, this application also provides a system for dynamic path deviation compensation of UAVs under strong wind conditions, comprising: a multi-source sensor fusion module for acquiring flight state information of the UAV and obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing; and a wind disturbance force estimation module, which has a built-in hybrid aerodynamic model for inputting the real-time three-dimensional wind field information into the hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV. The hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction. The outputs of the wind disturbance force and torque satisfy the following formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. The system comprises: a compensation force and torque output by the neural network compensation model; a dynamic inverse feedforward compensation module, used to generate feedforward compensation commands based on the wind disturbance force and torque, through a nonlinear dynamic inverse compensator considering actuator control margin constraints; a residual disturbance observation compensation module, used to estimate the residual disturbances that the hybrid aerodynamic model fails to characterize in real time using a high-order sliding mode disturbance observer, and feedforward the residual disturbance estimates to the feedforward compensation commands to obtain the corrected feedforward compensation commands; and a limiting and superposition control module, used to superimpose the corrected feedforward compensation commands, after adaptive limiting processing, with the output commands of the original UAV position controller and attitude controller to generate the final control signal for driving the actuators.

[0013] Furthermore, the wind disturbance force estimation model also includes an adaptive unscented Kalman filter, as a supplement to the extended Kalman filter, for online joint estimation of aerodynamic model parameters and wind field state; the adaptive unscented Kalman filter is activated when the wind speed change exceeds a preset threshold, and updates the process noise covariance matrix in real time according to the innovation sequence; the update law of the adaptive unscented Kalman filter satisfies: ,in for The process noise covariance matrix at time step 1. The forgetting factor has a value range of 0 to 1. for The process noise covariance matrix at time step 1. for The unscented Kalman filter gain matrix at time 1. for The information vector at time t, superscript This represents the transpose operation of a matrix or vector.

[0014] Furthermore, the system also includes a path dynamic replanning module, used to trigger path dynamic replanning when the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window. The path dynamic replanning is based on an improved fast random expansion tree algorithm, introducing an artificial potential field bias generated by the currently estimated wind field during random sampling and node expansion. The potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

[0015] The beneficial effects of this invention are as follows: It overcomes the measurement limitations of a single sensor by using multi-source sensor fusion processing, improving the estimation accuracy and reliability of real-time three-dimensional wind field information. It employs a hybrid aerodynamic model combining a physical mechanism base model and an online neural network compensation model in parallel, preserving the accuracy and interpretability of the physical model while possessing the adaptive learning capability of the neural network. This allows for real-time correction of model errors caused by aerodynamic parameter drift and unsteady aerodynamic effects. When generating feedforward compensation commands, it fully considers actuator control margin constraints and dynamically adjusts the compensation intensity based on the actuator's remaining control capability, fundamentally avoiding control failure caused by actuator saturation. It utilizes a high-order sliding mode disturbance observer to estimate residual disturbances that the hybrid aerodynamic model fails to characterize in real time and performs feedforward compensation, eliminating the impact of unmodeled dynamics and unknown disturbances on control accuracy. Finally, it generates the final control signal through adaptive amplitude limiting and command superposition, effectively suppressing measurement noise while ensuring the system's dynamic response speed. This comprehensively improves the path tracking accuracy, flight stability, and mission execution capability of the UAV in strong wind environments. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for dynamic path deviation compensation of a UAV in a strong wind environment, provided for an embodiment of this application;

[0017] Figure 2 This is a schematic diagram of a system for dynamic path deviation compensation of a UAV in a strong wind environment, provided as an embodiment of this application. Detailed Implementation

[0018] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0019] In this invention, the terms "system" and "network" are used interchangeably. "Multiple" refers to two or more; therefore, in this invention, "multiple" can also be understood as "at least two." "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / ", unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, it should be understood that in the description of this invention, terms such as "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or order.

[0020] Example 1:

[0021] like Figure 1 As shown in the embodiment of this application, a method for dynamic deviation compensation of UAV path under strong wind conditions is provided, including the following steps: S1, acquiring the flight status information of the UAV and obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing; S2, inputting the real-time three-dimensional wind field information into a hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV, wherein the hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction, and the output of the wind disturbance force and torque satisfies the formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. S3. Based on the wind disturbance force and torque, a feedforward compensation command is generated by a nonlinear dynamic inverse compensator that considers the actuator control margin constraint; S4. The residual disturbance that the hybrid aerodynamic model fails to characterize is estimated in real time using a high-order sliding mode disturbance observer, and the residual disturbance estimate is fed forward and superimposed into the feedforward compensation command to obtain a corrected feedforward compensation command; S5. The corrected feedforward compensation command is adaptively limited and then superimposed with the output commands of the original position controller and attitude controller of the UAV to generate the final control signal for driving the actuator.

[0022] In one possible embodiment, the flight status information of the UAV is first acquired, including basic flight parameters such as the UAV's three-dimensional position, three-dimensional velocity, attitude angle, angular velocity, rotor speed, and battery status. Then, multi-source sensor fusion processing is used to synchronize the flight status information and measurement data from different sensors in time, unify coordinates, and perform weighted fusion to eliminate measurement errors and delays from single sensors, obtaining accurate real-time three-dimensional wind field information. Next, the real-time three-dimensional wind field information is input into a hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV. The hybrid aerodynamic model consists of a base model based on physical mechanisms and a neural network for online correction. The network compensation models are constructed in parallel. The base model, based on physical mechanisms and incorporating the Newton-Euler rigid body dynamics equations, is established by considering the inherent parameters of the UAV's geometry, mass distribution, and rotor aerodynamic characteristics. This model can calculate the basic wind disturbance force and torque experienced by the UAV under ideal conditions. The neural network compensation model, through online learning of actual flight data, corrects errors in the base model caused by aerodynamic parameter drift and unsteady aerodynamic effects in real time. The total wind disturbance force and torque are the sum of the basic wind disturbance force and torque and the compensation force and torque. Then, based on the calculated wind disturbance force and torque, a nonlinear dynamic inverse compensator considering actuator control margin constraints generates feedforward compensation. The compensation command, based on the UAV dynamics model, converts the wind disturbance acceleration and angular acceleration to be offset into corresponding control inputs. During the conversion process, it acquires the current control output and maximum output capability of each actuator in real time, calculates the actuator control margin, and adaptively adjusts the compensation gain based on the control margin to ensure that the compensation command remains within the actuator's output capability range. Subsequently, a high-order sliding mode disturbance observer is used to estimate residual disturbances not represented by the hybrid aerodynamic model in real time. These residual disturbances include unsteady wind disturbances such as gusts and turbulence, as well as aerodynamic effects not modeled in the model. The high-order sliding mode disturbance observer employs an algorithm with finite-time convergence characteristics, enabling rapid... The residual disturbance is estimated quickly and accurately. The estimated residual disturbance value is then fed forward and superimposed onto the feedforward compensation command to obtain the corrected feedforward compensation command. Finally, the corrected feedforward compensation command is processed by adaptive amplitude limiting and superimposed on the output commands of the original position controller and attitude controller of the UAV to generate the final control signal for driving the actuator. The position controller generates position control command based on the deviation between the actual position of the UAV and the desired path, and the attitude controller generates attitude control command based on the deviation between the actual attitude of the UAV and the desired attitude. The adaptive amplitude limiting process dynamically adjusts the amplitude limiting threshold according to real-time wind field changes and flight status, suppressing measurement noise while ensuring the compensation effect.

[0023] By overcoming the measurement limitations of single sensors through multi-source sensor fusion processing, the estimation accuracy and reliability of real-time three-dimensional wind field information are improved. A hybrid aerodynamic model is adopted, which combines a physical mechanism basis model and an online neural network compensation model in parallel. This model retains the accuracy and interpretability of the physical model while possessing the adaptive learning capability of the neural network. It corrects model errors caused by aerodynamic parameter drift and unsteady aerodynamic effects in real time. When generating feedforward compensation commands, the actuator control margin constraints are fully considered, and the compensation intensity is dynamically adjusted according to the remaining control capability of the actuator, fundamentally avoiding control failure caused by actuator saturation. A high-order sliding mode disturbance observer is used to estimate residual disturbances that the hybrid aerodynamic model fails to represent in real time and perform feedforward compensation to eliminate the influence of unmodeled dynamics and unknown disturbances on control accuracy. Finally, the final control signal is generated by adaptive limiting and command superposition. While ensuring the dynamic response speed of the system, measurement noise is effectively suppressed, and the path tracking accuracy, flight stability and mission execution capability of UAVs in strong wind environments are comprehensively improved.

[0024] Existing wind field estimation methods typically use only a single wind speed sensor to measure local wind speed, which cannot accurately reflect the real wind field at the center of mass of the UAV. Furthermore, single filtering algorithms have poor tracking ability for abrupt wind field changes, resulting in significant delays and errors in wind field estimation, which cannot provide accurate input for subsequent wind disturbance compensation.

[0025] In this embodiment, the step S1, obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing, specifically includes: S11, measuring local wind speed vectors based on multiple wind speed sensors mounted on the UAV arm, performing coordinate transformation in conjunction with the UAV's real-time attitude angle, and obtaining the initial wind speed vector at the UAV's center of mass through weighted least squares spatial interpolation; S12, using accelerometer and satellite positioning velocity information as observations, performing error correction on the initial wind speed vector using an extended Kalman filter, and outputting optimal three-dimensional wind field information. The state update equation of the extended Kalman filter is: ,in For the first Time based on the past The optimal state estimation vector obtained from the observations For the first Time based on the past The prior state estimation vector obtained from the observations For the first The Kalman gain matrix at time t. For the first The observation vector at time t, To extend the nonlinear observation function of the Kalman filter, the state estimation vector includes three-dimensional wind speed components, UAV attitude angles, and angular velocity components.

[0026] In one possible embodiment, multiple wind speed sensors mounted on the drone's arms are first used to measure local wind speed vectors. These sensors, capable of simultaneously measuring three-dimensional wind speed components, are installed at the ends of each drone arm, away from the influence area of ​​the rotor downwash. Each wind speed sensor outputs a three-dimensional local wind speed vector at its corresponding location. Then, a coordinate transformation is performed using the drone's real-time attitude angles to convert the local wind speed vector in the aircraft's coordinate system to a unified ground coordinate system. The coordinate transformation uses a rotation matrix calculated from the drone's roll, pitch, and yaw angles to ensure that wind speed vectors measured at different locations are in the same coordinate system. Next, the initial wind speed vector at the drone's center of mass is obtained through weighted least squares spatial interpolation. The weights of the weighted least squares interpolation are determined based on the distance between each wind speed sensor and the drone's center of mass, as well as the confidence level of the sensor's historical measurements. The greater the distance, the higher the initial wind speed vector. The more stable the recent and historical measurement data of a sensor, the greater its weight. This results in an initial wind speed vector at the centroid that better reflects the overall wind conditions of the UAV. Finally, using accelerometer and satellite positioning velocity information as observations, an extended Kalman filter is used to correct the error of the initial wind speed vector, outputting the optimal three-dimensional wind field information. The state vector of the extended Kalman filter includes three-dimensional wind speed components, UAV attitude angular components, and angular velocity components. The observation vector includes three-dimensional acceleration measured by the accelerometer and three-dimensional velocity measured by the satellite positioning module. The extended Kalman filter first performs state prediction, predicting the prior state estimate of the current moment based on the optimal state estimate of the previous moment and the system dynamics model. Then, it performs observation update, calculating the Kalman gain based on the current moment's observations and correcting the prior state estimate to obtain the optimal state estimate of the current moment, thereby achieving high-precision, real-time estimation of the three-dimensional wind field.

[0027] By installing multiple wind speed sensors on the UAV arm to perform multi-point spatial measurements, and combining coordinate transformation and weighted least squares spatial interpolation, the initial wind speed vector at the centroid is obtained. This reduces the impact of fuselage obstruction and rotor downwash on wind speed measurement, and improves the spatial representativeness of local wind speed measurements. An extended Kalman filter is used with accelerometer and satellite positioning velocity information as observations to correct the initial wind speed vector error, achieving multi-state joint estimation. While correcting wind speed measurement errors, the attitude and angular velocity states of the UAV are estimated simultaneously, improving the real-time performance and accuracy of three-dimensional wind field estimation, and providing reliable basic data for subsequent wind disturbance compensation.

[0028] In this embodiment of the application, the neural network compensation model in S2 is an online updated radial basis function neural network. The radial basis function neural network takes the wind speed, attitude angle, rotor speed and the residual of the current mechanism model output and the force calculated based on sensor data within the historical time window as input, and outputs the compensation increment of the aerodynamic coefficient in real time, thereby obtaining the compensation force and torque. The online update of the radial basis function neural network adopts the recursive least squares algorithm to adjust the network weights.

[0029] In one possible embodiment, the radial basis function neural network consists of an input layer, a hidden layer, and an output layer. The input layer receives wind speed, attitude angle, rotor speed, and the residual between the current mechanistic model output and the force calculated from sensor data within a historical time window. The historical time window contains measurement data from multiple consecutive control cycles, reflecting the changing trends of aerodynamic parameters. The hidden layer contains multiple neurons, each using a Gaussian kernel function as the activation function. The relevant parameters of the Gaussian kernel function are predetermined during offline training. The output layer outputs the compensation increment of the aerodynamic coefficients, corresponding to the force coefficients and torque coefficients in three directions, respectively. The online update of the radial basis function neural network uses a recursive least squares algorithm to adjust the network weights. The recursive least squares algorithm updates the weight matrix of the output layer in real time by minimizing the squared error between the model output and the actual measured value. In each control cycle, the latest input data is first input into the neural network to obtain the aerodynamic coefficient compensation increment at the current moment. Then, based on the residual between the current mechanistic model output and the force calculated from sensor data, the recursive least squares algorithm is used to update the network weights, enabling the neural network to continuously adapt to changes in aerodynamic parameters and improve the prediction accuracy of the model.

[0030] By employing an online-updated radial basis function neural network as the compensation model, it features simple structure, low computational cost, and strong nonlinear approximation capability, making it suitable for the real-time operation requirements of UAV embedded systems. Using multi-dimensional flight data and model residuals within historical time windows as input, it comprehensively captures the influencing factors of aerodynamic parameter changes, accurately tracks aerodynamic parameter drift, and compensates for the output error of the physical mechanism basis model in real time, improving the overall accuracy and robustness of the hybrid aerodynamic model. The recursive least squares algorithm is used to adjust the network weights, resulting in fast convergence speed and high computational efficiency. It can complete weight updates within each control cycle, ensuring the real-time nature of the compensation.

[0031] In this embodiment of the application, the nonlinear dynamic inverse compensator considering actuator control margin constraints in step S3 acquires the control margin of each actuator in real time when solving for feedforward compensation acceleration and angular velocity, and processes it according to the following gain scheduling method: defining thrust margin: ;when At that time, the scaling factor of the vertical compensation acceleration gain is calculated according to the following formula: Otherwise, the scaling factor is 1; where This represents the actuator thrust margin, with a value ranging from 0 to 1. The maximum available thrust of a single rotor of the drone. For the first The current thrust command is constantly sent to the rotor. This is a preset critical margin threshold. This is the gain scaling factor for compensating acceleration in the vertical channel.

[0032] In one possible embodiment, the actuator thrust margin is first defined as the ratio of the remaining available thrust of the actuator to the maximum available thrust, reflecting the additional thrust capability that the actuator can still provide. For multi-rotor UAVs, the thrust margin of each rotor is calculated separately, and the overall system thrust margin is taken as the minimum value of the thrust margins of all rotors to ensure that none of the actuators enters a saturation state. Then, the vertical compensation acceleration gain is adaptively adjusted according to the thrust margin. When the thrust margin is greater than or equal to a preset critical margin threshold, it indicates that the actuator still has sufficient remaining control capability, and the vertical compensation acceleration gain scaling factor is 1, that is, no scaling is performed, according to the calculation... The obtained full compensation command is used for compensation. When the thrust margin is less than the critical margin threshold, it indicates that the actuator's remaining control capability is insufficient. The scaling factor of the vertical compensation acceleration gain is proportional to the thrust margin. The smaller the thrust margin, the smaller the scaling factor and the lower the compensation intensity, thus avoiding the compensation command from exceeding the actuator's maximum output range. The same adaptive scaling is performed on the roll, pitch, and yaw channels. The corresponding control margin is calculated for each channel, and the compensation angular acceleration gain of that channel is adjusted according to the control margin. Through the above gain scheduling method, it is ensured that the compensation command of each channel is always within the actuator's output capability range, avoiding control failure caused by actuator saturation.

[0033] By calculating the thrust margin of the actuator in real time and adaptively scaling the compensation acceleration gain of each channel, the actuator saturation is effectively avoided while ensuring the compensation effect. Adaptive scaling is performed on the roll, pitch and yaw channels respectively. The compensation intensity can be flexibly adjusted according to the actual control margin of each channel, making full use of the remaining control capability of the actuator, improving the robustness and safety of the system. The gain scheduling method is simple and efficient, with low computational load, and will not increase the burden on the system, thus meeting the requirements of real-time control.

[0034] In this embodiment of the application, the adaptive limiting processing in S5 employs a variable structure limiting filter, and the limiting threshold of the variable structure limiting filter is... The fuzzy logic regulator is based on the absolute value of the rate of change of wind speed. and position tracking error Real-time determination, expressed by the formula: When wind speed changes drastically and tracking error is large, the limiting threshold is adaptively increased to allow for large instantaneous compensation; conversely, the threshold is decreased to suppress measurement noise.

[0035] In one possible embodiment, the limiting threshold of the variable structure limiting filter is determined in real time by a fuzzy logic regulator. The inputs to the fuzzy logic regulator are the absolute value of the wind speed change rate and the position tracking error, and the output is the limiting threshold. First, the input and output variables are fuzzified, and the value range of each variable is divided into multiple fuzzy subsets. A corresponding membership function is defined for each fuzzy subset. Then, a fuzzy rule base is established. The fuzzy rules adopt the "IF-THEN" form and are formulated based on actual flight experience, covering all possible input combinations. Next, based on the current input variable value, the fuzzy value of the output variable is obtained through fuzzy inference. Finally, the centroid method is used for defuzzification to obtain the accurate limiting threshold. The variable structure limiting filter limits the corrected feedforward compensation command according to the limiting threshold. When the absolute value of the compensation command is less than or equal to the limiting threshold, the original compensation command is output. When the absolute value of the compensation command is greater than the limiting threshold, the limiting threshold is multiplied by the sign of the compensation command, thereby ensuring that the output command is within a reasonable range, taking into account both the dynamic performance and steady-state accuracy of the system.

[0036] The fuzzy logic controller determines the limiting threshold in real time based on the absolute value of the wind speed change rate and the position tracking error. It can adapt to different wind field environments and flight states. When the wind speed changes drastically and the tracking error is large, the limiting threshold is adaptively increased to allow for a larger instantaneous compensation amount and ensure the dynamic response speed of the system. When the wind speed is stable and the tracking error is small, the limiting threshold is adaptively decreased to suppress measurement noise and improve the steady-state accuracy of the system. The fuzzy logic controller does not require a precise mathematical model and can handle uncertainty and nonlinear problems well, with strong robustness.

[0037] In this embodiment of the application, the higher-order sliding mode perturbation observer in step S4 employs the super-twisting algorithm, and the residual perturbation estimation law of the higher-order sliding mode perturbation observer is: ;in, This is the estimated vector of the residual perturbation. The first design gain for the super-twisting algorithm, For sliding mode variables absolute value Exponentiation. For symbolic functions, This is an integration operation over time. This is the second design gain for the super-twisting algorithm. The sliding mode variable is calculated from the actual acceleration signal measured by the accelerometer and the wind disturbance force output by the hybrid aerodynamic model.

[0038] In one possible implementation, a sliding mode variable is first defined. This variable is calculated from the actual acceleration signal measured by the accelerometer and the wind disturbance force output by the hybrid aerodynamic model, reflecting the deviation between the actual and theoretical acceleration. Then, a super-twisting algorithm is used to design the residual perturbation estimation law. The algorithm consists of two parts. The first part is a proportional term, which is proportional to the product of the first power of the absolute value of the sliding mode variable and the sign function, and is used to accelerate the convergence speed. The second part is an integral term, which integrates the sign function of the sliding mode variable to eliminate steady-state error. By adjusting the values ​​of the two design gains, it can be ensured that the observer converges in a finite time and the estimation error converges to zero. The value of the design gain is determined according to the disturbance characteristics of the system. Under the premise of ensuring the stability of the observer, the convergence speed is improved as much as possible. During flight, the high-order sliding mode disturbance observer receives the actual acceleration signal measured by the accelerometer and the wind disturbance force output by the hybrid aerodynamic model in real time, calculates the sliding mode variable, and then calculates the estimated value of the residual disturbance in real time according to the residual disturbance estimation law. The estimated value is then fed forward and superimposed into the feedforward compensation command to compensate for the impact of the residual disturbance on the flight of the UAV.

[0039] By employing a high-order sliding mode disturbance observer based on the super-twisting algorithm, it possesses finite-time convergence characteristics and can accurately estimate various nonlinear and time-varying residual disturbances within a finite time. The estimation process does not depend on an accurate system model and exhibits strong robustness to system parameter variations and external disturbances. Furthermore, the algorithm has low computational cost, simple structure, and is suitable for the real-time operation requirements of UAV embedded systems. By compensating for residual disturbances through feedforward, it can significantly improve the system's anti-interference capability and control accuracy.

[0040] In this embodiment, after S5, the method further includes: S6, when the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window, triggering dynamic path replanning, wherein the dynamic path replanning is based on an improved fast random expansion tree algorithm, and an artificial potential field bias generated by the currently estimated wind field is introduced during random sampling and node expansion, wherein the potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

[0041] In one possible embodiment, the triggering condition for dynamic path replanning is first determined. When the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window, it indicates that the current wind disturbance compensation can no longer meet the path tracking requirements, and dynamic path replanning needs to be triggered. The preset threshold and preset time window are determined according to the task requirements and the flight performance of the UAV to avoid false triggering. Then, an improved fast random expansion tree algorithm is used for dynamic path replanning. An artificial potential field bias is introduced during random sampling and node expansion. The artificial potential field function consists of a target attraction potential field and a wind field guiding potential field. The target attraction potential field causes the sampling point to move closer to the target point, ensuring that the path can reach the target point. The wind field guiding potential field causes the sampling point to deflect in the downwind direction, thereby reducing the energy consumption of wind disturbance compensation. The fast random expansion tree algorithm first initializes the random... The tree is constructed with the UAV's current position as the root node. A point is randomly sampled within the flight space, and the potential field value of that point is calculated using an artificial potential field function. Points with lower potential field values ​​are more likely to be selected. The node closest to the sampled point is then found in the random tree, and a new node is generated by expanding the tree from that node towards the sampled point by one step. The new node is checked to see if it satisfies the flight constraints and avoids collisions with obstacles. If the new node is valid, it is added to the random tree, and the cost from the root node to the new node is calculated. Nodes near the new node are rewired to optimize the path cost of the random tree. This process is repeated until the random tree expands to the vicinity of the target point or the maximum number of iterations is reached. Finally, the path is backtracked from the target point to the root node to obtain the optimal path. The path is then smoothed to obtain a continuous and smooth new desired path, which is then output to the position controller for execution.

[0042] By introducing an artificial potential field bias generated by the currently estimated wind field during the path dynamic replanning process, the newly planned path tends to deflect towards the downwind direction, which can effectively reduce wind disturbance compensation energy consumption, extend the UAV's endurance, and improve mission execution efficiency. The improved fast randomized extended tree algorithm can quickly generate the optimal path that meets flight constraints, with good real-time performance, making it suitable for online path planning in dynamic environments. The artificial potential field function combines the target attraction potential field and the wind field guiding potential field, which can both ensure that the path moves towards the target point and make full use of the wind field energy to achieve global path optimization.

[0043] Example 2:

[0044] Reference Figure 2This application also provides a system for dynamic path deviation compensation of a UAV under strong wind conditions, including: a multi-source sensor fusion module for acquiring flight status information of the UAV and obtaining real-time three-dimensional wind field information through multi-source sensor fusion processing; and a wind disturbance force estimation module, which has a built-in hybrid aerodynamic model for inputting the real-time three-dimensional wind field information into the hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV. The hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction. The output of the wind disturbance force and torque satisfies the following formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. The system comprises: a compensation force and torque output by the neural network compensation model; a dynamic inverse feedforward compensation module, used to generate feedforward compensation commands based on the wind disturbance force and torque, through a nonlinear dynamic inverse compensator considering actuator control margin constraints; a residual disturbance observation compensation module, used to estimate the residual disturbances that the hybrid aerodynamic model fails to characterize in real time using a high-order sliding mode disturbance observer, and feedforward the residual disturbance estimates to the feedforward compensation commands to obtain the corrected feedforward compensation commands; and a limiting and superposition control module, used to superimpose the corrected feedforward compensation commands, after adaptive limiting processing, with the output commands of the original UAV position controller and attitude controller to generate the final control signal for driving the actuators.

[0045] In one possible embodiment, the system for dynamic path deviation compensation of UAVs in strong wind environments includes a multi-source sensor fusion module, a wind disturbance force estimation module, a dynamic inverse feedforward compensation module, a residual disturbance observation and compensation module, and a limiting and superposition control module. The multi-source sensor fusion module is connected to various sensors such as wind speed sensors, accelerometers, satellite positioning modules, and inertial measurement units, and receives measurement data from each sensor in real time. It performs time synchronization, coordinate transformation, and weighted fusion processing on the measurement data to obtain real-time three-dimensional wind field information, which is then sent to the wind disturbance force estimation module via a high-speed bus. The wind disturbance force estimation module has a built-in hybrid aerodynamic model. It receives the real-time three-dimensional wind field information from the multi-source sensor fusion module, combines it with the UAV's flight status information, calculates the total wind disturbance force and torque experienced by the UAV, and simultaneously sends the total wind disturbance force and torque to the dynamic inverse feedforward compensation module and the residual disturbance observation and compensation module. The dynamic inverse feedforward compensation module receives the wind disturbance force and torque from the wind disturbance force estimation module and simultaneously acquires data from each sensor in real time. The system calculates the actuator control margin based on the current control output and maximum output capability of the actuator. It then generates a feedforward compensation command using a nonlinear dynamic inverse algorithm that considers the actuator control margin constraint. This command is sent to the residual disturbance observation and compensation module. The residual disturbance observation and compensation module receives the feedforward compensation command from the dynamic inverse feedforward compensation module and simultaneously uses a high-order sliding mode disturbance observer to estimate residual disturbances that the hybrid aerodynamic model fails to characterize in real time. The residual disturbance estimates are then fed forward and superimposed onto the feedforward compensation command to obtain a corrected feedforward compensation command. This command is then sent to the adaptive limiting and superposition control module. The adaptive limiting and superposition control module receives the corrected feedforward compensation command from the residual disturbance observation and compensation module and simultaneously receives control commands output from the position controller and attitude controller. After adaptive limiting processing, the feedforward compensation command is superimposed with the output commands from the position controller and attitude controller to generate the final motor control signal, which is then sent to the UAV's motor drive module to drive the UAV to perform corresponding flight maneuvers.

[0046] In this embodiment, the wind disturbance estimation model further includes an adaptive unscented Kalman filter as a supplement to the extended Kalman filter, used for online joint estimation of aerodynamic model parameters and wind field state; the adaptive unscented Kalman filter is activated when the wind speed change exceeds a preset threshold, and updates the process noise covariance matrix in real time according to the innovation sequence; the update law of the adaptive unscented Kalman filter satisfies: ,in for The process noise covariance matrix at time step 1. The forgetting factor has a value range of 0 to 1. for The process noise covariance matrix at time step 1. for The unscented Kalman filter gain matrix at time 1. for The information vector at time t, superscript This represents the transpose operation of a matrix or vector.

[0047] In one possible embodiment, an adaptive unscented Kalman filter is used for online joint estimation of aerodynamic model parameters and wind field conditions. The system monitors the rate of change of wind speed in real time. When the rate of change of wind speed exceeds a preset threshold, it indicates a sudden change in the wind field, and the system automatically switches from an extended Kalman filter to an adaptive unscented Kalman filter. The adaptive unscented Kalman filter uses a symmetric sampling method to generate sigma points, which are then used to... The point propagation approximates the probability distribution of the state, avoiding the calculation of the Jacobian matrix in the extended Kalman filter and improving the estimation accuracy of nonlinear systems. At the same time, the adaptive unscented Kalman filter updates the process noise covariance matrix in real time according to the innovation sequence. The innovation sequence is the difference between the observed and the observed prediction, reflecting the deviation between the model prediction and the actual measurement. The update law of the process noise covariance matrix adopts the form of exponential weighted average. The weight of historical data is adjusted by the forgetting factor, which enables the filter to quickly track the changes in system characteristics. The value of the forgetting factor is determined according to the dynamic characteristics of the system, balancing the tracking ability and smoothing ability of the filter. When the wind speed change rate is lower than the preset threshold and continues for a certain period of time, it indicates that the wind field has returned to stability, and the system automatically switches back to the extended Kalman filter to reduce the amount of computation and improve the operating efficiency of the system.

[0048] In this embodiment, the method further includes a path dynamic replanning module, used to trigger path dynamic replanning when the horizontal position deviation of the UAV after compensation and superposition exceeds a preset threshold and the duration exceeds a preset time window. The path dynamic replanning is based on an improved fast random expansion tree algorithm, which introduces an artificial potential field bias generated by the currently estimated wind field during random sampling and node expansion. The potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

[0049] In one possible embodiment, the path dynamic replanning module is connected to the amplitude limiting and overlay control module and the position controller. The adaptive amplitude limiting and overlay control module sends the actual flight position information of the UAV to the path dynamic replanning module in real time. The position controller sends the desired path information to the path dynamic replanning module. The path dynamic replanning module compares the deviation between the actual flight position and the desired path in real time, calculates the magnitude and duration of the horizontal position deviation, and automatically activates the path replanning function when the horizontal position deviation exceeds a preset threshold and the duration exceeds a preset time window. The path dynamic replanning module uses the wind field potential field guided RRT algorithm to generate a new desired path online. The wind field potential field guided RRT algorithm introduces an artificial potential field bias generated by the currently estimated wind field on the basis of the traditional RRT* algorithm, so that the new planned path has a tendency to deflect in the downwind direction, reducing the energy consumption of wind disturbance compensation. After generating the new desired path, the path dynamic replanning module directly outputs the path to the position controller. The position controller immediately adjusts the control command according to the new desired path to realize automatic path tracking and complete the closed-loop control of the entire process from deviation detection to path replanning and then to path tracking.

[0050] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0051] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0052] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A method for dynamic path deviation compensation of unmanned aerial vehicles (UAVs) under strong wind conditions, characterized in that, Includes the following steps: S1. Acquire the flight status information of the UAV and obtain real-time three-dimensional wind field information through multi-source sensor fusion processing; S2. Input the real-time three-dimensional wind field information into the hybrid aerodynamic model to calculate the wind disturbance force and torque experienced by the UAV. The hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction. The output of the wind disturbance force and torque satisfies the following formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. The compensation force and torque output by the neural network compensation model; S3. Based on the wind disturbance force and torque, a feedforward compensation command is generated by a nonlinear dynamic inverse compensator that considers the actuator control margin constraint. S4. Use a high-order sliding mode disturbance observer to estimate the residual disturbances that the hybrid aerodynamic model fails to characterize in real time, and feed forward the residual disturbance estimates to the feedforward compensation command to obtain the corrected feedforward compensation command. S5. The modified feedforward compensation command is processed by adaptive amplitude limiting and then superimposed with the output commands of the original position controller and attitude controller of the UAV to generate the final control signal for driving the actuator.

2. The method for dynamic deviation compensation of UAV path under strong wind conditions according to claim 1, characterized in that, The real-time three-dimensional wind field information obtained through multi-source sensor fusion processing in S1 specifically includes: S11. Based on the measurement of local wind speed vector by multiple wind speed sensors installed on the drone arm, coordinate transformation is performed in combination with the real-time attitude angle of the drone, and the initial wind speed vector at the center of mass of the drone is obtained by weighted least squares spatial interpolation. S12. Using accelerometer and satellite positioning velocity information as observations, an extended Kalman filter is used to correct the error of the initial wind speed vector, outputting the optimal three-dimensional wind field information. The state update equation of the extended Kalman filter is: ,in For the first Time based on the past The optimal state estimation vector obtained from the observations For the first Time based on the past The prior state estimation vector obtained from the observations For the first The Kalman gain matrix at time t. For the first The observation vector at time t, To extend the nonlinear observation function of the Kalman filter, the state estimation vector includes three-dimensional wind speed components, UAV attitude angles, and angular velocity components.

3. The method for dynamic path deviation compensation of UAVs under strong wind conditions according to claim 1, characterized in that, The neural network compensation model in S2 is an online-updated radial basis function neural network. The radial basis function neural network takes the wind speed, attitude angle, rotor speed and the residual of the current mechanism model output and the force calculated based on sensor data within the historical time window as input, and outputs the compensation increment of the aerodynamic coefficient in real time, thereby obtaining the compensation force and torque. The online update of the radial basis function neural network adopts the recursive least squares algorithm to adjust the network weights.

4. The method for dynamic deviation compensation of UAV path under strong wind conditions according to claim 1, characterized in that, The nonlinear dynamic inverse compensator in S3, which considers actuator control margin constraints, acquires the control margin of each actuator in real time when solving for feedforward compensation acceleration and angular velocity, and processes it according to the following gain scheduling method: Define thrust margin: ; when At that time, the scaling factor of the vertical compensation acceleration gain is calculated according to the following formula: Otherwise, the scaling factor is 1; in This represents the actuator thrust margin, with a value ranging from 0 to 1. The maximum available thrust of a single rotor of the drone. For the first The current thrust command is constantly sent to the rotor. This is a preset critical margin threshold. This is the gain scaling factor for compensating acceleration in the vertical channel.

5. The method for dynamic path deviation compensation of UAVs under strong wind conditions according to claim 1, characterized in that, The adaptive limiting process in S5 employs a variable structure limiting filter, and the limiting threshold of the variable structure limiting filter is... The fuzzy logic regulator is based on the absolute value of the rate of change of wind speed. and position tracking error Real-time determination, expressed by the formula: ; When wind speed changes drastically and tracking error is large, the limiting threshold is adaptively increased to allow for large instantaneous compensation; conversely, the threshold is decreased to suppress measurement noise.

6. The method for dynamic deviation compensation of UAV path under strong wind conditions according to claim 1, characterized in that, The higher-order sliding mode perturbation observer in S4 employs the super-twisting algorithm, and the residual perturbation estimation law of the higher-order sliding mode perturbation observer is: ;in, This is the estimated vector of the residual perturbation. The first design gain for the super-twisting algorithm, For sliding mode variables absolute value Exponentiation. For symbolic functions, This is an integration operation over time. This is the second design gain for the super-twisting algorithm. The sliding mode variable is calculated from the actual acceleration signal measured by the accelerometer and the wind disturbance force output by the hybrid aerodynamic model.

7. The method for dynamic deviation compensation of UAV path under strong wind conditions according to claim 1, characterized in that, Following S5 are: S6. When the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window, dynamic path replanning is triggered. This dynamic path replanning is based on an improved fast random expansion tree algorithm. During random sampling and node expansion, an artificial potential field bias generated by the currently estimated wind field is introduced. The potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

8. A system for dynamic path deviation compensation of unmanned aerial vehicles (UAVs) under strong wind conditions, characterized in that, The method for compensating for dynamic deviations in the path of a UAV under strong wind conditions, as described in any one of claims 1-7, includes: The multi-source sensor fusion module is used to acquire the flight status information of the UAV and obtain real-time three-dimensional wind field information through multi-source sensor fusion processing. The wind disturbance force estimation module has a built-in hybrid aerodynamic model. This model receives the real-time 3D wind field information and calculates the wind disturbance force and torque experienced by the UAV. The hybrid aerodynamic model includes a parallel base model based on physical mechanisms and a neural network compensation model for online correction. The outputs of the wind disturbance force and torque satisfy the following formula: ,in The total wind disturbance force and torque experienced by the drone. The basic wind disturbance force and moment output by the base model. The compensation force and torque output by the neural network compensation model; The dynamic inverse feedforward compensation module is used to generate feedforward compensation commands based on the wind disturbance force and torque, through a nonlinear dynamic inverse compensator that takes into account the actuator control margin constraint. The residual disturbance observation and compensation module is used to estimate the residual disturbances that the hybrid aerodynamic model fails to characterize in real time using a high-order sliding mode disturbance observer, and feed forward the residual disturbance estimates to the feedforward compensation command to obtain the corrected feedforward compensation command. The limiting and superposition control module is used to superimpose the modified feedforward compensation command, after adaptive limiting processing, with the output commands of the original position controller and attitude controller of the UAV to generate the final control signal for driving the actuator.

9. The system for dynamic path deviation compensation of UAVs under strong wind conditions according to claim 8, characterized in that, The wind disturbance estimation model also includes an adaptive unscented Kalman filter for online joint estimation of aerodynamic model parameters and wind field state. The adaptive unscented Kalman filter is activated when a sudden change in wind speed exceeds a preset threshold, and updates the process noise covariance matrix in real time based on the innovation sequence. The update law of the adaptive unscented Kalman filter satisfies: ,in for The process noise covariance matrix at time step 1. The forgetting factor has a value range of 0 to 1. for The process noise covariance matrix at time step 1. for The unscented Kalman filter gain matrix at time 1. for The information vector at time t, superscript This represents the transpose operation of a matrix or vector.

10. The system for dynamic path deviation compensation of UAVs under strong wind conditions according to claim 8, characterized in that, The system also includes a path dynamic replanning module, which triggers path dynamic replanning when the horizontal position deviation of the UAV after compensation exceeds a preset threshold and the duration exceeds a preset time window. The path dynamic replanning is based on an improved fast random expansion tree algorithm, introducing an artificial potential field bias generated by the currently estimated wind field during random sampling and node expansion. The potential field function of the artificial potential field is defined as: ;in, Sampling points The artificial potential field value at that location, The attraction coefficient of the target point to the sampling point. For Euclidean norm operations. This is a vector of random sampling point locations during the path planning process. Let the target point position vector be the UAV flight mission. This represents the guiding coefficient of the wind field on the sampling point. This is the currently estimated three-dimensional wind speed vector. This is the current position vector of the drone.

Images