Dual-mode dynamic cooperative control method and system based on pulling of mooring line

Abstract

The invention discloses a dual-mode dynamic cooperative control method and system based on pulling of a mooring line. The method comprises the steps that displacement electric signals are collected through double potentiometers; a Z-axis PWM signal is generated based on the displacement electric signal through a PWM signal generator; an XY-axis PWM signal is generated through the included angle difference between an XY rocker on the unmanned aerial vehicle and the mooring line; the Z-axis PWM signal is divided into two paths, one path is transmitted to a capstan electronic speed controller, and the other path is transmitted to the unmanned aerial vehicle and is combined with the X-axis PWM signal and the Y-axis PWM signal to form a three-dimensional control signal; setting a remote control signal and a winch potentiometer signal as well as a three-dimensional control signal and a winch manual signal as a first linkage group and a second linkage group respectively; when the first linkage group is switched on and the second linkage group is switched off, the unmanned aerial vehicle enters a first control mode; and when the second linkage group is switched on and the first linkage group is switched off, the unmanned aerial vehicle enters a second control mode. The method is used for solving the technical problems of single control mode, insufficient man-machine collaboration, poor environmental interference adaptability and limited control precision existing in traditional mooring unmanned aerial vehicle control.

Description

Technical Field

[0001] The present invention relates to the technical field of tethered drone control, and in particular to a dual-mode dynamic collaborative control method and system based on pulling a tether line. Background Art

[0002] Traditional tethered drone control technology generally suffers from problems such as a single control mode, insufficient human-machine collaboration, and poor environmental adaptability. These problems are specifically manifested in the following aspects: 1. Limitations of Control Mode and Operational Collaboration Existing technologies often rely on remote controls to independently control the drone and the winch's retraction and release lines. This requires manual synchronization of the aircraft's attitude and the tension of the tethering lines, creating complex operations and prone to safety risks due to response delays. For example, traditional systems lack a dual-mode linkage mechanism, preventing flexible switching between conventional remote control scenarios and close-range precision tracking. This results in insufficient control accuracy and operational convenience for specific operations, such as emergency manual intervention and high-precision trajectory tracking.

[0003] 2. Interference between environment and mechanical characteristics Temperature sensitivity: The elastic modulus of mooring line materials fluctuates significantly with temperature (for example, the elastic modulus of some materials decreases by 2%-5% for every 10°C increase in temperature). Traditional fixed-parameter models can easily cause Z-axis control signal deviations exceeding 15% under different temperature environments (-20°C to 60°C), leading to tension loss or decreased lifting accuracy.

[0004] Light and swing interference: Horizontal position detection relies on a single sensor (such as a visual or mechanical angle sensor) and is susceptible to environmental interference: The visual sensor's deviation in extracted pixel coordinates of the mooring line increases in strong light overexposure or low light noise scenes. The high-frequency oscillation (0.5-5Hz) and elastic deformation of the mooring line of the mechanical displacement sensor cause nonlinear distortion of the angle difference data. Traditional linear fitting models are difficult to compensate for such errors.

[0005] 3. Sensor Data Processing Defects Signal noise and drift: Traditional tension detection uses a single potentiometer to collect displacement signals, which is easily affected by changes in mechanical clearance and contact resistance, resulting in a low signal-to-noise ratio. The rocker arm pulley assembly is installed on the drone end, and the change in weight caused by changes in the length of the tethering line causes the potentiometer median value to drift.

[0006] 4. Control Coupling and Dynamic Response Lag The traditional system does not consider the dynamic coupling effect between the deployment length of the tether line and the speed of the UAV, resulting in cross-interference between the XY-axis horizontal control and the Z-axis vertical control; there is a lack of a feedforward prediction mechanism, and the position response lags behind the changes in wind disturbance, resulting in trajectory deviation; and there is no multi-modal protection logic based on the swing frequency. In resonance scenarios (such as when the swing frequency exceeds 3Hz), it is easy to cause attitude instability or cable entanglement risks.

[0007] The above problems result in poor adaptability and high operating threshold of traditional tethered drones in complex working conditions, making it difficult to meet the needs of high-precision operations. Therefore, there is an urgent need for a dynamic control solution that takes into account control flexibility, environmental robustness and collaborative accuracy. Summary of the Invention

[0008] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a dual-mode dynamic collaborative control method and system based on pulling a tethered line, which is used to solve the technical problems existing in the control of traditional tethered drones, such as a single control mode, insufficient human-machine collaboration, poor adaptability to environmental interference and limited control accuracy.

[0009] In order to solve the above problems, the technical solutions adopted by the present invention are as follows: A dual-mode dynamic cooperative control method based on pulling a mooring line comprises the following steps: The displacement electrical signal is collected through a double potentiometer configured on the rocker pulley assembly of the tethered ground equipment; The PWM signal generator generates a dynamic elastic modulus based on the tether line elastic coefficient model using the real-time collected tether line ambient temperature data, and combines it with the displacement electrical signal to generate a Z-axis PWM signal representing the pulling; The angle difference between the XY joystick on the drone and the tether line is used to generate an XY axis PWM signal representing the horizontal position based on the angle difference-coordinate mapping model and dynamic filtering. The Z-axis PWM signal is divided into two paths. One path is transmitted to the winch electronic controller to control the reeling and releasing speed. The other path is transmitted to the drone and combined with the XY-axis PWM signal to form a three-dimensional control signal. Set the remote control signal and the winch potentiometer signal as the first linkage group, and set the three-dimensional control signal and the winch manual signal as the second linkage group; When the first linkage group is turned on and the second linkage group is disconnected through the first and second remote control relays, the drone enters the first control mode; when the second linkage group is turned on and the first linkage group is disconnected, the drone enters the second control mode.

[0010] Preferably, when entering the first control mode, the drone receives the remote control signal, and the winch electric control receives the winch potentiometer signal, and performs remote control of the drone's XYZ three-axis direction, and superimposes the tension fluctuation suppression factor on the Z-axis PWM signal to automatically reverse the winch motor to reel in and out the line; When entering the second control mode, the drone receives three-dimensional control signals, and the winch electronic speed controller receives manual winch signals. When the winch is manually reeling in the line, the PWM signal generator outputs a PWM signal representing the descending Z-axis. When the winch is manually releasing the line, the PWM signal generator outputs a PWM signal representing the ascending Z-axis. Combined with the XY-axis PWM signals, a new three-dimensional control signal is generated, and the drone's XYZ three-axis follow-by-wire control is performed based on nonlinear decoupling.

[0011] Preferably, when generating the XY axis PWM signal, the method includes: The binocular vision sensor carried by the UAV collects the projection image of the mooring line on the horizontal plane, and extracts the pixel coordinates of the mooring line from the projection image; The analog signal of the joystick deflection angle is obtained through the mechanical displacement sensor of the XY joystick and converted into angle difference data; Based on the ambient light intensity and swing frequency, the pixel coordinates and angle difference data are corrected respectively, and the angle difference-coordinate mapping model constructed based on the BP neural network is input to output the predicted XY coordinates; The Kalman filter algorithm is used to perform real-time error compensation on the predicted XY coordinates and generate XY axis PWM signals; Among them, the state covariance matrix P of the filtering process is dynamically adjusted according to the real-time error.

[0012] Preferably, when correcting pixel coordinates based on ambient light intensity, the method includes: Continuously collect multiple ambient light intensity samples and divide them into three intervals based on intensity: low light, medium light, and high light. A logarithmic enhancement algorithm is used to improve pixel contrast in the low light interval, and dynamic range compression is used to suppress overexposed areas in the high light interval to obtain processed ambient light intensity samples. The sample variance of the processed ambient light intensity samples is calculated. When the variance is less than or equal to the set value, the mean is directly taken as the current light intensity. When the variance is greater than the set value, the median filter is started, and the mean is taken as the current light intensity after removing the outliers. The preset interval compensation coefficient is queried according to the current light intensity, and the pixel coordinates are corrected by nonlinear mapping based on the logarithmic operation result of the current light intensity and the standard reference light intensity.

[0013] Preferably, when correcting the angle difference data based on the swing frequency, the method includes: Obtain the natural frequency curves of the mooring line under different tensions and establish a swing frequency-stiffness mapping model; The current swing frequency of the mooring line is collected in real time, and the current equivalent stiffness coefficient is calculated through the swing frequency-stiffness mapping model. It is then compared with the standard stiffness coefficient to generate the stiffness correction coefficient λ; Multiply the angle difference data by the stiffness correction coefficient λ, and introduce the swing phase compensation term for correction; The modified angle difference data is denoised using a variable step-size LMS adaptive filter, and the step-size factor μ is dynamically adjusted according to the swing frequency.

[0014] Preferably, when outputting the predicted XY coordinates, the following steps are included: The corrected pixel coordinates are assigned a dynamic weight based on the interval compensation coefficient through the input layer, and the corrected angle difference data is assigned a dynamic weight based on the stiffness correction coefficient λ, and the weight distribution ratio is optimized in real time through the gradient descent algorithm; The Leaky ReLU activation function is used from the input layer to the first hidden layer to process the pixel coordinate features, and the ELU activation function is used from the second hidden layer to the output layer to process the angle difference features. A Batch Normalization layer is added to the output layer to suppress gradient diffusion. It also includes: the steps of constructing an adversarial training sample set based on Gaussian noise, random scaling and generative adversarial networks, as well as the steps of online iterative optimization of the model based on coordinate error calculation, Bayesian optimization algorithm update and dynamic adjustment of learning rate.

[0015] Preferably, when performing real-time error compensation, the method includes: When the root mean square error between the XY coordinates measured by the GPS of the drone and the XY coordinates output by the filter is greater than the set value for multiple consecutive sampling cycles, the adaptive optimization of the Kalman gain K is started: The root mean square error is divided into three levels: low error, medium error and high error through the fuzzy logic controller, and the corresponding adjustment coefficient K value is assigned; when the error level jumps from low and medium error to high error, the strong tracking factor t is triggered to perform weighted correction on the state covariance matrix P of the Kalman filter.

[0016] Preferably, when the tension fluctuation suppression factor is superimposed, it includes: The high-frequency fluctuation component of the mooring line tension is collected and decomposed into three frequency bands through wavelet transform; Adaptive threshold noise reduction is performed on each frequency band component to retain the effective frequency band that represents the elastic vibration of the cable; The noise-reduced fluctuation component is converted into an inverse compensation PWM signal and linearly superimposed with the Z-axis PWM signal; The amplitude of the reverse compensation PWM signal is positively correlated with the root mean square value of the tension fluctuation, and the phase lag is set to 90° to achieve active vibration reduction.

[0017] Preferably, when performing the XYZ three-axis line control of the drone, the following steps are included: The XY-axis PWM signal and the Z-axis PWM signal are nonlinearly decoupled using the coupling coefficient matrix of the tethered line's deployment length and the drone's real-time speed. The coupling coefficient matrix is ​​dynamically updated based on the tethered line's elastic coefficient model. An inverse UAV dynamics model is introduced to predict the UAV position deviation based on the current three-dimensional control signal. A feedforward compensation is generated and superimposed on the control signal. The compensation amplitude is positively correlated with the first-order derivative of the tether line tension fluctuation. When the tether line swing frequency exceeds the set value, the XY axis control weight is reduced and the attitude angle limit protection is activated. At the same time, the Z axis control is switched to constant tension priority mode, and the tension fluctuation is controlled through the PID parameter self-tuning algorithm.

[0018] A dual-mode dynamic cooperative control system based on pulling a mooring line, when in operation, executes the above method, including: Tethered ground equipment: includes a rocker arm pulley assembly, a tethering line, a tethering box winch, and a PWM signal generator. The rocker arm pulley assembly is equipped with a dual potentiometer for collecting displacement electrical signals. The PWM signal generator generates a Z-axis PWM signal based on the displacement electrical signal and divides it into two channels: one for transmission to the winch ESC and the other for transmission to the UAV. Drone: Includes flight control system, XY joystick, and signal merging module. The XY joystick generates XY-axis PWM signals based on the angle difference between the joystick and the tethered line. The signal merging module receives the Z-axis PWM signal sent by the tethered ground device and merges it with the XY-axis PWM signal to form a three-dimensional control signal. Switching execution unit: includes the first and second remote control relays and the remote control; the first and second remote control relays are each equipped with two sets of linkage contacts; the first remote control relay is used to control the control signal received by the flight control system, and the second remote control relay is used to control the control signal received by the winch electronic controller; the remote control is used to send the remote control signal of the drone and to control the conduction or disconnection of the remote control relay to make the drone enter the first or second control mode.

[0019] Compared with the prior art, the present invention has the following beneficial effects: (1) Dual-mode collaborative control to improve operational flexibility By dynamically switching between the first control mode (conventional remote control with automatic retraction and release) and the second control mode (follow-by-wire control with manual intervention), the system balances automated operations with high-precision manual control. For example, the second control mode allows the drone to follow the aircraft in all directions by pulling the tethered line, while the retraction and landing phase achieves a fully synchronized landing, significantly simplifying operation.

[0020] (2) Adaptive correction of environment and mechanical characteristics to improve control accuracy Dynamic temperature compensation: The elastic modulus of the mooring line is corrected in real time based on the ambient temperature, solving the problem of excessive tension detection error at different temperatures in traditional fixed parameter models.

[0021] Light and swing interference suppression: By fusing binocular vision with mechanical displacement sensor data, combined with a light interval enhancement algorithm (low-light logarithmic enhancement, high-light dynamic compression) and a swing frequency-stiffness mapping model, horizontal position detection errors are reduced.

[0022] (3) Multi-source data fusion and dynamic modeling to enhance system robustness A BP neural network is used to construct an angle difference-coordinate nonlinear mapping model, and the Kalman filter is combined to dynamically adjust the state covariance matrix to achieve real-time error compensation of the predicted coordinates. The variable step-size LMS adaptive filtering and wavelet transform active vibration reduction technology are introduced to effectively suppress the high-frequency vibration of the mooring line and ensure signal stability in complex environments.

[0023] (4) Nonlinear decoupling and feedforward control to optimize dynamic response Nonlinear decoupling of the X, Y, and Z axes is achieved through the coupling coefficient matrix of the tether line deployment length and the drone speed. A dynamic inverse model is introduced to predict position deviation and generate feedforward compensation to shorten the trajectory tracking lag time. A multi-modal protection logic triggered by the swing frequency is designed (for example, the constant tension priority mode is activated when it is greater than 3 Hz) to avoid the risk of attitude instability caused by resonance.

[0024] In summary, the present invention significantly improves the operational convenience, control accuracy, and environmental adaptability of tethered UAVs through innovative control modes, environmental adaptive algorithms, and dynamic collaborative mechanisms, and can be widely used in various high-precision operation scenarios.

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a step diagram of a dual-mode dynamic cooperative control method according to an embodiment of the present invention; Figure 2 This is a structural diagram of a rocker arm pulley assembly according to an embodiment of the present invention; Figure 3 is a flow chart of generating XY axis PWM signals according to an embodiment of the present invention; Figure 4 is a flow chart of correcting pixel coordinates based on ambient light intensity according to an embodiment of the present invention; Figure 5 is a flow chart of correcting angle difference data based on swing frequency according to an embodiment of the present invention; Figure 6 It is a module interaction diagram of a dual-mode dynamic collaborative control system according to an embodiment of the present invention.

[0027] Explanation of the accompanying figures: 20. Dual-mode dynamic cooperative control system; 21. Tethered ground equipment; 211. PWM signal generator; 212. Rocker arm pulley assembly; 213. Dual potentiometer; 214. Winch electric regulator; 215. Winch motor; 22. UAV; 221. Flight control system; 222. XY joystick; 223. Signal merging module; 224. Receiver; 231. First remote control relay; 232. Second remote control relay; 233. Remote control. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0029] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0030] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0031] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0032] Example 1, see Figure 1 The present invention provides a dual-mode dynamic cooperative control method step diagram. Figure 1 A dual-mode dynamic cooperative control method based on pulling a tethered line is shown, comprising the following steps: S101. Pulling state detection: The relative pulling state between the UAV 22 and the tethering line is detected by the rocker pulley assembly 212 on the tethered ground device 21; wherein, the dual potentiometer 213 configured by the rocker pulley assembly 212 collects the displacement electrical signal, and generates a Z-axis PWM signal representing the pulling through the PWM signal generator 211; S102 horizontal state detection: the angle difference between the XY rocker 222 and the tethered line on the drone 22 is used to generate an XY axis PWM signal representing the horizontal position; S103. Signal distribution and transmission: The Z-axis PWM signal is divided into two paths, one of which is transmitted to the winch ESC 214 to control the retracting and releasing speed, and the other is transmitted to the drone 22 via wired or wireless means and combined with the XY-axis PWM signal to form a three-dimensional control signal. S104 signal synchronization logic: the remote control signal and the capstan potentiometer signal is set to the first linkage group, the three-dimensional control signal and the capstan manual signal is set to the second linkage group; S105. Control mode switching: When the first linkage group is turned on by the first remote control relay 231 and the second remote control relay 232 and the second linkage group is disconnected, the drone 22 enters the first control mode; when the second linkage group is turned on by the first remote control relay 231 and the second remote control relay 232 and the first linkage group is disconnected, the drone 22 enters the second control mode; The first remote control relay 231 is provided on the UAV 22 , and the second remote control relay 232 is provided on the tethered ground equipment 21 .

[0033] See Figure 2 FIG2 is a structural diagram of a rocker pulley assembly. It should be further explained in the embodiment of the present invention that the pull-sensing rocker pulley assembly 212 is deployed on the tethered ground equipment 21 in this embodiment, which solves the problem of the potentiometer median value drift caused by the weight change due to the change in the length of the tethering line when the rocker pulley assembly 212 is installed on the drone 22 (the horizontal XY axis control detection function is still integrated in the drone 22).

[0034] Background description: In the control of traditional tethered drones 22, there are common problems such as a single control mode and insufficient operational coordination. For example, existing technologies mostly rely on remote controls 233 to achieve independent control of the drone 22 and the winch reel and release line, requiring manual and synchronous adjustment of the aircraft's attitude and the tension of the tether line. The operation is highly complex and prone to safety risks due to response delays. At the same time, in specific operating scenarios (such as close-range precision following and emergency manual intervention), traditional remote control modes are difficult to balance control accuracy and operational flexibility, and there are problems such as excessive tension fluctuations and horizontal following lags. Based on this: In the above step S105, when entering the first control mode, the drone 22 receives the remote control signal, and the winch electric controller 214 receives the winch potentiometer signal to perform remote control of the X, Y, and Z axes of the drone 22, and a tension fluctuation suppression factor is superimposed on the Z-axis PWM signal to automatically rotate the winch motor 215 in the forward and reverse directions to reel in and out the line. When entering the second control mode, the drone 22 receives a three-dimensional control signal, and the winch ESC 214 receives a manual winch signal. When the winch is manually reeling in the line, the PWM signal generator 211 outputs a Z-axis PWM signal representing the descent. When the winch is manually releasing the line, the PWM signal generator 211 outputs a Z-axis PWM signal representing the rise. Combined with the XY-axis PWM signal, a new three-dimensional control signal is generated, and the drone 22 performs wire-controlled follow-up of the XYZ three axes based on nonlinear decoupling.

[0035] What needs to be further explained in the embodiment of the present invention is that this embodiment realizes the coordinated operation of the UAV 22 and the tethered ground equipment 21 by designing a dual-mode linkage control logic: the first control mode meets the requirements of automatic line retraction and release in conventional remote control scenarios, and the second control mode achieves high-precision following and emergency response capabilities through dynamic coupling of manual operation and line control signals, thereby improving the reliability and operational convenience of the tethered UAV 22 in various operating scenarios.

[0036] The second control mode realizes the height control of the UAV 22 by pulling the tethering line, and can complete all-round following of the horizontal XY axis and the vertical Z axis during the following process of the tethered UAV 22, and achieve fully synchronized landing during the line-reeling and landing phase, which greatly reduces the difficulty of the landing operation of the tethered UAV 22.

[0037] Background: Traditional tethered drone tension detection schemes generally use a fixed elastic modulus parameter for tension-to-displacement conversion, without considering the impact of ambient temperature on the physical properties of the tether line. The elastic modulus of tether line materials (such as high-strength fibers and composite cables) fluctuates significantly with temperature (for example, for every 10°C increase in temperature, the elastic modulus of some materials may decrease by 2%-5%), leading to systematic errors in tension detection. Directly calculating tension using the static elastic modulus at standard temperature can cause deviations exceeding 15% in the Z-axis PWM signal output in extreme temperature environments (such as -20°C or 60°C), leading to a significant decrease in the drone's lift control accuracy or the risk of uncontrolled tether line tension.

[0038] At the same time, traditional tension detection relies on a single potentiometer to collect displacement signals, which is easily interfered by factors such as mechanical gaps and contact resistance changes, resulting in high signal noise. In addition, if the temperature sensor is installed on the UAV 22 end, it will increase the aircraft load and affect the temperature measurement stability due to airflow disturbances. However, temperature collection on the ground equipment end can improve the accuracy and reliability of ambient temperature monitoring through fixed installation. Based on this: In the above step S101, when generating the Z-axis PWM signal representing the pulling, it includes: Obtain the displacement electrical signal collected by the dual potentiometer 213, generate the dynamic elastic modulus through the elastic coefficient model of the tethered line, and generate the Z-axis PWM signal by combining the displacement electrical signal and the dynamic elastic modulus; When generating the dynamic elastic modulus, the method includes: constructing an elastic coefficient model of the mooring line based on the elastic modulus at the standard temperature, the temperature coefficient of the mooring line material, the standard temperature, and the ambient temperature of the mooring line; The real-time collected mooring line ambient temperature data is input into the mooring line elastic coefficient model to generate the difference with the standard temperature. The elastic modulus at the standard temperature is corrected in combination with the temperature coefficient of the mooring line material to output the dynamic elastic modulus. The ambient temperature of the mooring line is obtained by a temperature sensor provided on the mooring ground equipment 21 .

[0039] What needs to be further explained in the embodiment of the present invention is that this embodiment proposes a dynamic elastic modulus correction mechanism: the ambient temperature is collected in real time by a ground temperature sensor, an elastic coefficient model is constructed in combination with the material temperature coefficient, and the elastic modulus parameters are dynamically corrected to ensure the linearity of the tension-displacement conversion under different temperature conditions; at the same time, a dual potentiometer 213 is used to differentially collect the displacement signal to reduce the risk of single-point failure, improve the signal-to-noise ratio of the original signal, and provide a reliable data basis for the accurate generation of the Z-axis PWM signal.

[0040] Background description: In traditional tethered UAV22 horizontal position detection schemes, a single sensor (such as pure vision or pure mechanical angle sensor) is often used for position estimation, which has problems such as poor environmental adaptability and large model linearization error. For example, visual sensors are easily affected by changes in light intensity (such as strong light overexposure and low-light noise), resulting in deviations in the extraction of pixel coordinates of the tethered line; mechanical displacement sensors produce angle difference data drift due to factors such as tethered line swing and material elastic deformation. In addition, traditional models often use linear fitting methods to establish the mapping relationship between sensor data and position, which makes it difficult to deal with nonlinear factors such as light interference and mechanical lag, resulting in insufficient output accuracy of the horizontal control signal (XY axis PWM), especially under complex working conditions (such as strong wind disturbances and rapid attitude adjustments), which are prone to follow-up delays or overshoots. Based on this: See Figure 3 The flowchart of generating an XY axis PWM signal, in the above step S102, when generating an XY axis PWM signal representing a horizontal position, includes: The binocular vision sensor carried by the UAV 22 collects a projection image of the mooring line on the horizontal plane, and extracts the pixel coordinates of the mooring line from the projection image; The mechanical displacement sensor of the XY rocker 222 obtains an analog signal of the rocker deflection angle and converts it into angle difference data; After correcting the pixel coordinates and angle difference data, they are used as model input; The model input is input into the angle difference-coordinate mapping model built based on the BP neural network, and the predicted XY coordinates are output; The Kalman filter algorithm is used to perform real-time error compensation on the predicted XY coordinates and generate XY axis PWM signals; Among them, the state covariance matrix P of the filtering process is dynamically adjusted according to the real-time error.

[0041] What needs to be further explained in the embodiment of the present invention is that this embodiment proposes a multi-source data fusion and dynamic modeling method: through the complementary heterogeneous data of binocular vision and mechanical displacement sensors, combined with the ambient light and swing frequency correction mechanism, the reliability of the original data is improved; the BP neural network is used to construct a nonlinear mapping model to achieve high-precision conversion of angle difference-coordinates; and the Kalman filter is introduced to dynamically adjust the state covariance matrix, compensate for the prediction error in real time, and finally generate a more robust XY-axis PWM signal to meet the horizontal position control requirements of the tethered drone 22 in a complex environment.

[0042] In a possible embodiment, correcting pixel coordinates and angle difference data includes: The ambient light intensity is collected in real time by using a light sensor module integrated into the drone 22; The oscillation frequency of the mooring line is collected by a miniature triaxial acceleration sensor installed at the anchor point at the bottom of the mooring line; Correct pixel coordinates based on ambient light intensity, and correct angle difference data based on swing frequency; The light sensor module uses a silicon-based photodiode array with a range of 0-100,000 Lux and a sampling frequency of ≥10Hz. 2 The C bus communicates with the flight control system 221; the miniature three-axis acceleration sensor has a built-in MEMS vibration detection unit with a sampling frequency of ≥100 Hz, which can output the X / Y / Z axis acceleration components in real time, convert the time domain vibration signal into frequency domain data through Fourier transform, and extract the main swing frequency as the tether line swing frequency.

[0043] What needs to be further explained in the embodiment of the present invention is that this embodiment proposes a multi-source environmental parameter fusion correction strategy: the high-precision light sensor module (silicon-based photodiode array) carried by the drone 22 monitors the light intensity in real time, and dynamically corrects the pixel coordinates in combination with the interval compensation algorithm; at the same time, a miniature three-axis acceleration sensor is deployed at the anchor point at the bottom of the mooring line, and the swing frequency is collected using the MEMS vibration detection unit. The angle difference data is optimized through the stiffness mapping model and adaptive filtering. In terms of sensor selection, the light module covers a range of 0-100000 Lux and supports I 2 C high-speed communication, the acceleration sensor sampling frequency ≥ 100Hz and the ability to perform Fourier transform frequency domain analysis ensure the real-time and accuracy of environmental parameter collection, and provide high-quality input data for subsequent BP neural network coordinate mapping.

[0044] Background description: In the traditional tethered UAV 22 visual positioning system, changes in light intensity will cause the contrast and edge clarity of the tethered line projection image to drop significantly. For example, in low-light environments, the increase in image noise leads to deviations in pixel coordinate extraction, and in high-light environments, overexposed areas cause the tethered line outline to be lost. Traditional solutions mostly use fixed image enhancement algorithms (such as global contrast stretching) and do not perform interval processing for the dynamic range of light (0-100000 Lux), resulting in poor correction effects under different lighting conditions. At the same time, the sampling data of the light sensor is susceptible to instantaneous interference (such as cloud occlusion and ground reflection), resulting in jumps. Traditional mean filtering is difficult to effectively eliminate outliers, further exacerbating coordinate mapping errors. In addition, the mapping relationship between pixel coordinates and actual physical coordinates is nonlinear, and traditional linear interpolation correction cannot compensate for the complex coupling relationship between light intensity and imaging distortion, resulting in reduced horizontal position control accuracy. Based on this: See Figure 4 A flowchart of correcting pixel coordinates, in one possible embodiment, correcting pixel coordinates based on ambient light intensity includes: Light intensity interval division: The light sensor module continuously collects 10 ambient light intensity samples and divides them into three intervals based on intensity: low light (0-1000 Lux), medium light (1001-50000 Lux), and high light (50001-100000 Lux). A logarithmic enhancement algorithm is used to improve pixel contrast in the low light interval, and dynamic range compression is used to suppress overexposed areas in the high light interval to obtain the processed ambient light intensity samples. Real-time noise suppression: Calculate the sample variance of the processed ambient light intensity samples. When the variance is less than or equal to 500 Lux, 2 When the variance is greater than 500 Lux, the mean is taken as the current light intensity. 2 When , median filtering is started, outliers are removed and the mean value is taken as the current light intensity; Coordinate mapping correction: Query the preset interval compensation coefficient according to the current light intensity, and perform nonlinear mapping correction on the pixel coordinates based on the logarithmic operation result of the current light intensity and the standard reference light intensity.

[0045] What needs to be further explained in the embodiment of the present invention is that this embodiment proposes an adaptive illumination correction mechanism: by dividing the illumination interval into three levels and configuring corresponding processing strategies (low-light logarithmic enhancement, high-light dynamic compression), image quality optimization under full-range illumination is achieved; a median filtering algorithm triggered by a variance threshold is adopted to improve the stability of illumination sampling data; a nonlinear mapping model combining interval compensation coefficients and logarithmic operations is used to accurately correct the pixel coordinate offset caused by illumination changes, providing high signal-to-noise ratio input data for subsequent BP neural network coordinate prediction, thereby improving the environmental robustness of horizontal position detection of the tethered drone 22.

[0046] Background description: In the mechanical displacement sensing system of a traditional tethered UAV 22, changes in the tether line's swing frequency will cause dynamic fluctuations in the equivalent stiffness coefficient. However, the traditional solution does not establish a correlation model between the swing frequency and mechanical characteristics. Instead, it directly uses a fixed stiffness coefficient to calculate the angle difference data. As a result, when the tension changes (such as tension fluctuations of ±30% during the lifting and lowering of the UAV 22) or external disturbances (such as 0.5-5Hz swing frequency changes caused by gusts), the stiffness coefficient error can reach 15%-20%, causing nonlinear distortion of the angle difference data.

[0047] At the same time, the mooring line is prone to resonance in the 0.5-2Hz range. The output signal of the traditional mechanical displacement sensor has a significant phase lag (the lag can reach 10°-15°), and the filter parameters are not dynamically adjusted for different swing frequencies. The fixed-step LMS filter has a slow convergence speed in the high-frequency band (convergence time >200ms when >5Hz) and insufficient filtering accuracy in the low-frequency band (noise suppression ratio <20dB when <1Hz), further exacerbating the measurement error of the angle difference data. Based on this: See Figure 5 A flowchart of correcting angle difference data, in a possible embodiment, correcting the angle difference data based on the swing frequency includes: Construction of the swing frequency-stiffness mapping model: Obtain the natural frequency curve of the mooring line under different tensions and establish a nonlinear mapping model between the swing frequency and the equivalent stiffness coefficient, where the equivalent stiffness coefficient K=ω 2 m / L (ω is the oscillation frequency, m is the mass per unit length, and L is the length of the mooring line); Dynamic correction coefficient generation: The current swing frequency of the mooring line is collected in real time, and the current equivalent stiffness coefficient is calculated through the swing frequency-stiffness mapping model. This is compared with the standard stiffness coefficient to generate the stiffness correction coefficient λ = K / K0 (K0 is the stiffness coefficient under the standard state of 25°C); Mechanical displacement compensation: The angle difference data is multiplied by the stiffness correction coefficient λ, and a swing phase compensation term is introduced for correction. When the swing frequency is in the resonant range of 0.5-2Hz, a phase lag compensation based on the inverse tangent function is superimposed, and the compensation amplitude is positively correlated with the swing amplitude. Adaptive filtering optimization: A variable step-size LMS adaptive filter is used to reduce the noise of the corrected angle difference data. The step-size factor μ is dynamically adjusted according to the swing frequency. In the high-frequency band (>5Hz), the μ value is 0.01-0.05 to speed up convergence. In the medium-frequency band (1-5Hz), the μ value is 0.005-0.01 to balance the convergence speed and filtering accuracy. In the low-frequency band (<1Hz), the μ value is 0.001-0.005 to improve the filtering accuracy.

[0048] What needs to be further explained in the embodiment of the present invention is that this embodiment proposes an oscillation frequency coupling correction mechanism: by establishing an oscillation frequency-stiffness mapping model to calculate the equivalent stiffness coefficient in real time, a dynamic correction coefficient λ is generated to compensate for the stiffness change; an inverse tangent function phase compensation term is introduced for the resonance range to offset the influence of mechanical lag; and a variable step-size LMS adaptive filter is used (high frequency band μ=0.01-0.05, medium frequency band μ=0.005-0.01, low frequency band μ=0.001-0.005) to achieve full-band noise adaptive suppression, thereby effectively controlling the angle difference data measurement error and providing high-precision input for XY-axis PWM signal generation.

[0049] Background: Traditional angle difference-to-coordinate mapping models use a fixed weight distribution strategy when fusing multi-source data. This strategy fails to dynamically adjust feature contributions based on sensor data reliability. For example, when drastic changes in illumination cause a surge in pixel coordinate noise, fixed weights introduce redundant features, reducing model robustness. Furthermore, a single activation function (such as pure ReLU or Sigmoid) struggles to balance the linear distribution of pixel coordinates with the nonlinear characteristics of angle difference data. This results in insufficient hidden layer feature representation and deep networks are prone to gradient vanishing (gradient descent attenuation >60%), impacting the real-time performance of coordinate mapping.

[0050] In addition, traditional training sample sets are mostly collected in laboratory environments and lack data under extreme working conditions (such as a sudden increase in the stiffness of the mooring line due to low temperatures of -20°C, and high-frequency oscillation caused by strong winds of 15m / s), which limits the generalization ability of the model. Moreover, after offline training is completed, the parameters are fixed and cannot cope with sensor drift during long-term flight (such as error accumulation caused by temperature drift of MEMS devices). The deviation between the predicted coordinates and the actual GPS measured values ​​gradually increases over time. Based on this: In a possible embodiment, outputting the predicted XY coordinates includes: Adaptive weight distribution is performed through the input layer of the angle difference-coordinate mapping model: the corrected pixel coordinates are assigned a dynamic weight based on the interval compensation coefficient, and the corrected angle difference data is assigned a dynamic weight based on the stiffness correction coefficient λ. The weight distribution ratio is optimized in real time using a gradient descent algorithm to reduce feature redundancy under extreme environmental interference. The hidden layers of the angle difference-coordinate mapping model use a hybrid activation function combination for processing: the Leaky ReLU activation function is used from the input layer to the first hidden layer to process pixel coordinate features, and the ELU activation function is used from the second hidden layer to the output layer to process angle difference features. A batch normalization layer is added to the output layer to suppress gradient diffusion and improve the real-time convergence speed of coordinate mapping. This also includes the steps of building an adversarial training sample set and online iterative optimization of the model; The adversarial training sample set was constructed by adding Gaussian noise (mean 0, variance 0.02-0.1) to the original pixel coordinate samples to simulate camera shake. The angle difference samples were randomly scaled (scaling factor 0.8-1.2) to simulate elastic deformation of the mooring line. Virtual samples under extreme working conditions were generated using a generative adversarial network (GAN), ensuring that the training set covered the temperature range of -20°C to 60°C and wind speeds of 10-15 m / s. Model online iterative optimization steps: Every preset flight cycle (10-30 seconds), the error between the drone's 22GPS measured XY coordinates and the angle difference-coordinate mapping model output is calculated, and the network connection weights and bias terms are updated using the Bayesian optimization algorithm. The learning rate is dynamically adjusted during the iteration process. When the error is less than 0.5 meters, the learning rate is reduced to 1 / 5 of the initial value to stabilize the model output.

[0051] What needs to be further explained in the embodiments of the present invention is that this embodiment proposes a dynamic modeling and online optimization mechanism: weight distribution (pixel coordinate weight associated with illumination compensation coefficient, angle difference weight associated with stiffness correction coefficient λ) is optimized in real time through gradient descent to achieve adaptive feature screening; Leaky ReLU-ELU hybrid activation function and BatchNormalization are used to suppress gradient diffusion and improve feature extraction efficiency; a GAN virtual sample set is constructed to cover a wide temperature range (-20°C to 60°C) and strong wind scenes, and the network parameters are iteratively updated every 10-30 seconds through Bayesian optimization (the learning rate is reduced to 1 / 5 when the error is <0.5 meters), to stably control the prediction error of the XY coordinates and provide high-precision initial input for the Kalman filter.

[0052] Background: In traditional tethered drone navigation systems, Kalman filters often use a fixed gain matrix, making it impossible to dynamically adjust state estimation weights based on real-time errors. When the deviation between the measured GPS coordinates and the predicted values ​​suddenly changes (e.g., due to electromagnetic interference causing the RMSE to temporarily exceed 0.3 meters), the fixed gain causes filter lag, and the state covariance matrix P cannot converge quickly, causing position tracking errors to accumulate. Furthermore, traditional error classification relies heavily on empirical thresholds and lacks the continuous domain partitioning capabilities of fuzzy logic. This makes it difficult to quantify the nonlinear impact of different error levels (low / medium / high) on gain adjustment, resulting in insufficient system robustness. Based on this: In a possible embodiment, performing real-time error compensation includes: A prediction error feedback mechanism is introduced: when the root mean square error (RMSE) between the XY coordinates measured by the GPS of the drone 22 and the XY coordinates output by the filter is greater than 0.3 meters for three consecutive sampling periods, the adaptive optimization of the Kalman gain K is started: The root mean square error (RMSE) is divided into three levels: low error (<0.1m), medium error (0.1-0.3m), and high error (>0.3m) through the fuzzy logic controller, and the corresponding K value adjustment coefficients are assigned, which are 0.9, 1.0, and 1.2 respectively; When the error level jumps from low error or medium error to high error, the strong tracking factor t (value ranges from 1.5 to 3.0) is triggered to perform weighted correction on the state covariance matrix P of the Kalman filter to enhance the tracking ability of sudden error.

[0053] What needs to be further explained in the embodiments of the present invention is that this embodiment dynamically divides the error levels and matches the Kalman gain coefficients (low error 0.9, medium error 1.0, high error 1.2) through a fuzzy logic controller to achieve continuous domain adaptive adjustment of the gain; introduces a strong tracking factor (1.5-3.0) to perform weighted correction on the state covariance matrix (P) to enhance the ability to quickly track sudden errors.

[0054] Background: In traditional winch retractable / retractable control systems, Z-axis PWM signal generation fails to account for high-frequency fluctuations in mooring line tension (elastic vibrations above 10 Hz), and tension feedback is achieved solely through simple PID closed-loop control. This lack of frequency-domain decomposition of the tension signal prevents differentiation between the cable's inherent elastic vibrations (10-50 Hz) and external wind disturbances (<10 Hz). This results in a phase difference between the motor's output torque and the actual required tension (a lag of up to 45°-60°). In strong winds of 15 m / s, tension fluctuations can reach ±20%, causing cable slack or over-tensioning, resulting in UAV 22 attitude oscillations (pitch angle fluctuations of ±3°).

[0055] At the same time, traditional filtering methods (such as sliding average filtering) have limited effectiveness in suppressing multi-band mixed noise, cannot effectively extract the key 10-50Hz vibration component, and lack reverse compensation accuracy. In addition, the compensation signal and the original PWM signal are often superimposed in phase, which can only achieve passive vibration reduction and cannot offset the elastic vibration energy with a 90° phase difference. As a result, the control accuracy of the retractable cable is difficult to meet the requirements of the UAV 22 fixed-point hovering. Based on this: In a possible embodiment, when the tension fluctuation suppression factor is superimposed, it includes: The high-frequency fluctuation component (above 10 Hz) of the mooring line tension is collected and decomposed into three frequency bands through wavelet transform; Adaptive threshold noise reduction is performed on each frequency band component to retain the effective frequency band that represents the elastic vibration of the cable; The noise-reduced fluctuation component is converted into an inverse compensation PWM signal and linearly superimposed with the Z-axis PWM signal to offset the influence of high-frequency tension oscillation on the accuracy of take-up and pay-off. The amplitude of the reverse compensation PWM signal is positively correlated with the root mean square value of the tension fluctuation, and the phase lag is set to 90° to achieve active vibration reduction.

[0056] The present embodiment requires further explanation regarding the active vibration reduction control mechanism proposed in this embodiment. This mechanism uses a wavelet transform to decompose tension fluctuations into three frequency bands. After applying adaptive threshold noise reduction, the effective vibration component is extracted, generating a reverse compensation PWM signal with a 90° phase lag. This is linearly superimposed with the Z-axis control signal to achieve active vibration reduction. The compensation signal amplitude is dynamically correlated with the root mean square value of the tension fluctuations, increasing the suppression rate of high-frequency vibrations (above 10 Hz), thereby effectively controlling the amplitude of the tension fluctuations in the pay-out and retract lines and significantly improving the vertical position control accuracy of the UAV 22.

[0057] Background description: In traditional three-axis CNC following systems for tethered UAVs 22, linear decoupling control strategies are often used. This strategy does not consider the dynamic coupling effect between the deployed length of the tethering line and the real-time speed of the UAV 22. This results in cross-interference between the XY-axis horizontal control and the Z-axis vertical control. Especially when the tethering line elastically deforms, the control signals interfere with each other, causing the trajectory of the UAV 22 to deviate.

[0058] At the same time, traditional feedback control relies on hysteresis adjustment of position error and lacks a feedforward prediction mechanism. When UAV 22 encounters sudden wind disturbances, the position response lags, causing the tether line tension to overshoot. In addition, the system lacks multi-modal switching logic. When the tether line swing frequency exceeds 3Hz (such as resonance caused by strong winds), it still maintains conventional control parameters, causing attitude instability and even the risk of cable entanglement. Based on this: In a possible embodiment, when performing three-axis XYZ follow-by-wire control of the drone 22, the following steps are included: Dynamic coupling compensation: The XY-axis PWM signal and the Z-axis PWM signal are nonlinearly decoupled through the coupling coefficient matrix of the tether line deployment length and the real-time speed of the UAV 22. The coupling coefficient matrix is ​​dynamically updated based on the tether line elastic coefficient model. Predictive feedforward control: Introducing an inverse dynamic model of the drone 22, the position deviation of the drone 22 in 0.1-0.3 seconds is predicted based on the current three-dimensional control signal. A feedforward compensation is generated and superimposed on the control signal. The compensation amplitude is positively correlated with the first-order derivative of the mooring line tension fluctuation. Multi-mode switching logic: When the mooring line swing frequency exceeds 3Hz, the emergency follow mode is automatically activated: at this time, the XY axis control weight is reduced by 20% and the attitude angle limit protection is activated. At the same time, the Z axis control switches to constant tension priority mode, and the PID parameter self-tuning algorithm is used to control the tension fluctuation within the rated value range of ±5%.

[0059] What needs to be further explained in the embodiments of the present invention is that this embodiment proposes a dynamic collaborative control mechanism: the coupling coefficient matrix is ​​updated in real time through the tethered line elastic coefficient model to achieve nonlinear decoupling of the XY-Z axes; a dynamic inverse model is introduced to predict the 0.1-0.3 second position deviation to generate a feedforward compensation associated with the first-order derivative of the tension fluctuation; a multi-modal switching logic based on the swing frequency is designed (>3Hz activates the emergency mode), the control weight is dynamically adjusted (XY axis is reduced by 20%) and constant tension priority control is started (fluctuation ±5% of the rated value), so as to reduce the three-axis following error and significantly improve the trajectory tracking accuracy under complex working conditions.

[0060] Example 2, see Figure 6 The dual-mode dynamic cooperative control system module interaction diagram, the present invention provides the following Figure 6 A dual-mode dynamic cooperative control system 20 based on pulling a tethered line is shown, comprising: a tethered ground device 21, a drone 22 and a switching execution unit.

[0061] The tethered ground equipment 21 includes a rocker arm pulley assembly 212, a tethering line, a tethering box winch and a PWM signal generator 211; among them, the rocker arm pulley assembly 212 is equipped with a dual potentiometer 213 for converting mechanical displacement into a displacement electrical signal; the PWM signal generator 211 is used to generate a Z-axis PWM signal representing the pulling according to the displacement electrical signal, and is divided into two paths, one is transmitted to the winch electric regulator 214 to control the reeling and releasing speed, and the other is transmitted to the drone 22 via wired or wireless means.

[0062] The drone 22 includes a flight control system 221, an XY rocker 222, and a signal merging module 223. The XY rocker 222 is used to generate an XY-axis PWM signal representing the horizontal position based on the angle difference between the rocker and the tethering line. The signal merging module 223 is used to receive the Z-axis PWM signal sent by the tethered ground device 21 and merge it with the XY-axis PWM signal to form a three-dimensional control signal. The switching execution unit includes a first remote control relay 231, a second remote control relay 232 and a remote controller 233; the first remote control relay 231 and the second remote control relay 232 are each configured with two sets of linkage contacts; the first remote control relay 231 is used to control the control signal received by the flight control system 221, and the second remote control relay 232 is used to control the control signal received by the winch electronic controller 214; the remote controller 233 is used to send remote control signals for the UAV 22 and to enable the UAV 22 to enter the first control mode or the second control mode by controlling the conduction or disconnection of the remote control relay.

[0063] In a possible embodiment, the mooring box winch further includes: a winch potentiometer and a winch motor 215; When entering the first control mode, the first remote control relay 231 controls the flight control system 221 to receive the remote control signal, while the second remote control relay 232 controls the winch electric controller 214 to receive the winch potentiometer signal, the UAV 22 performs XYZ three-axis remote control, and the winch motor 215 automatically reels in and out in the forward and reverse directions; When entering the second control mode, the first remote control relay 231 controls the flight control system 221 to receive the three-dimensional control signal. At the same time, the second remote control relay 232 controls the winch electronic controller 214 to receive the winch manual signal. When the winch is manually reeling in the line, the PWM signal generator 211 outputs a Z-axis PWM signal representing the descent. When the winch is manually releasing the line, the PWM signal generator 211 outputs a Z-axis PWM signal representing the rise. Combined with the XY axis PWM signal, a new three-dimensional control signal is generated, and the drone 22 performs wire-controlled follow-up on the XYZ three axes.

[0064] In a possible embodiment, the drone 22 further includes a receiver 224; the receiver 224 is used to receive the drone 22 remote control signal and the remote control relay control signal sent by the remote controller 233 to control the flight of the drone 22 and the on and off of the remote control relay.

[0065] In a possible embodiment, the mooring ground equipment 21 is further equipped with a temperature sensor for collecting the ambient temperature of the mooring line.

[0066] In a possible embodiment, the drone 22 is further configured with a binocular vision sensor for collecting a projection image of the mooring line on a horizontal plane and extracting pixel coordinates of the mooring line from the projection image.

[0067] In a possible embodiment, the XY rocker 222 is further configured with a mechanical displacement sensor for collecting analog signals of the rocker deflection angle and converting the signals into angle difference data.

[0068] In a possible embodiment, the drone 22 is further equipped with a light sensor module for real-time acquisition of ambient light intensity. Preferably, the light sensor module uses a silicon-based photodiode array with a range of 0-100,000 Lux and a sampling frequency of ≥10 Hz, and is connected to the I 2 The C bus communicates with the flight control system 221 .

[0069] In one possible embodiment, a miniature triaxial accelerometer is configured at the bottom anchor point of the mooring line to detect the mooring line's oscillation frequency. Preferably, the miniature triaxial accelerometer has a built-in MEMS vibration detection unit with a sampling frequency ≥ 100 Hz, capable of real-time output of X / Y / Z-axis acceleration components. The time-domain vibration signal is converted into frequency-domain data through Fourier transform, and the dominant oscillation frequency is extracted as the mooring line's oscillation frequency.

[0070] In a possible embodiment, the mechanical displacement sensor is configured with a variable step-size LMS adaptive filter for performing noise reduction processing on the corrected angle difference data.

[0071] In one possible embodiment, the XY joystick 222 is further configured with a fuzzy logic controller for dividing the root mean square error (RMSE) into three levels: low error (<0.1m), medium error (0.1-0.3m), and high error (>0.3m), and assigning adjustment coefficients of corresponding K values ​​of 0.9, 1.0, and 1.2, respectively.

[0072] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dual-mode dynamic cooperative control method based on pulling a mooring line, characterized in that: The following steps are involved: The displacement electrical signal is collected through a double potentiometer configured on the rocker pulley assembly of the tethered ground equipment; The PWM signal generator generates a dynamic elastic modulus based on the tether line elastic coefficient model using the real-time collected tether line ambient temperature data, and combines it with the displacement electrical signal to generate a Z-axis PWM signal representing the pulling; The angle difference between the XY joystick on the drone and the tether line is used to generate an XY axis PWM signal representing the horizontal position based on the angle difference-coordinate mapping model and dynamic filtering. The Z-axis PWM signal is divided into two paths. One path is transmitted to the winch electronic controller to control the reeling and releasing speed. The other path is transmitted to the drone and combined with the XY-axis PWM signal to form a three-dimensional control signal. Set the remote control signal and the winch potentiometer signal as the first linkage group, and set the three-dimensional control signal and the winch manual signal as the second linkage group; When the first linkage group is turned on and the second linkage group is disconnected through the first and second remote control relays, the drone enters the first control mode; when the second linkage group is turned on and the first linkage group is disconnected, the drone enters the second control mode.

2. The method according to claim 1, characterized in that When entering the first control mode, the drone receives the remote control signal, and the winch electric control receives the winch potentiometer signal at the same time, performing remote control of the drone's XYZ three-axis direction, and superimposing the tension fluctuation suppression factor on the Z-axis PWM signal to automatically reverse the winch motor to reel in and out the line; When entering the second control mode, the drone receives three-dimensional control signals, and the winch electronic speed controller receives manual winch signals. When the winch is manually reeling in the line, the PWM signal generator outputs a PWM signal representing the descending Z-axis. When the winch is manually releasing the line, the PWM signal generator outputs a PWM signal representing the ascending Z-axis. Combined with the XY-axis PWM signals, a new three-dimensional control signal is generated, and the drone's XYZ three-axis follow-by-wire control is performed based on nonlinear decoupling.

3. The method according to claim 1, characterized in that When generating XY axis PWM signals, it includes: The binocular vision sensor carried by the UAV collects the projection image of the mooring line on the horizontal plane, and extracts the pixel coordinates of the mooring line from the projection image; The analog signal of the joystick deflection angle is obtained through the mechanical displacement sensor of the XY joystick and converted into angle difference data; Based on the ambient light intensity and swing frequency, the pixel coordinates and angle difference data are corrected respectively, and the angle difference-coordinate mapping model constructed based on the BP neural network is input to output the predicted XY coordinates; The Kalman filter algorithm is used to perform real-time error compensation on the predicted XY coordinates and generate XY axis PWM signals; Among them, the state covariance matrix P of the filtering process is dynamically adjusted according to the real-time error.

4. The method according to claim 3, characterized in that When correcting pixel coordinates based on ambient light intensity, this includes: Continuously collect multiple ambient light intensity samples and divide them into three intervals based on intensity: low light, medium light, and high light. A logarithmic enhancement algorithm is used to improve pixel contrast in the low light interval, and dynamic range compression is used to suppress overexposed areas in the high light interval to obtain processed ambient light intensity samples. The sample variance of the processed ambient light intensity samples is calculated. When the variance is less than or equal to the set value, the mean is directly taken as the current light intensity. When the variance is greater than the set value, the median filter is started, and the mean is taken as the current light intensity after removing the outliers. The preset interval compensation coefficient is queried according to the current light intensity, and the pixel coordinates are corrected by nonlinear mapping based on the logarithmic operation result of the current light intensity and the standard reference light intensity.

5. The method according to claim 3, characterized in that When correcting the angle difference data based on the swing frequency, it includes: Obtain the natural frequency curves of the mooring line under different tensions and establish a swing frequency-stiffness mapping model; The current swing frequency of the mooring line is collected in real time, and the current equivalent stiffness coefficient is calculated through the swing frequency-stiffness mapping model. It is then compared with the standard stiffness coefficient to generate the stiffness correction coefficient λ; Multiply the angle difference data by the stiffness correction coefficient λ, and introduce the swing phase compensation term for correction; The modified angle difference data is denoised using a variable step-size LMS adaptive filter, and the step-size factor μ is dynamically adjusted according to the swing frequency.

6. The method according to claim 3, characterized in that When outputting predicted XY coordinates, this includes: The corrected pixel coordinates are assigned a dynamic weight based on the interval compensation coefficient through the input layer, and the corrected angle difference data is assigned a dynamic weight based on the stiffness correction coefficient λ, and the weight distribution ratio is optimized in real time through the gradient descent algorithm; The Leaky ReLU activation function is used from the input layer to the first hidden layer to process the pixel coordinate features, and the ELU activation function is used from the second hidden layer to the output layer to process the angle difference features. A Batch Normalization layer is added to the output layer to suppress gradient diffusion. It also includes: the steps of constructing an adversarial training sample set based on Gaussian noise, random scaling and generative adversarial networks, as well as the steps of online iterative optimization of the model based on coordinate error calculation, Bayesian optimization algorithm update and dynamic adjustment of learning rate.

7. The method according to claim 3, characterized in that When performing real-time error compensation, it includes: When the root mean square error between the XY coordinates measured by the GPS of the drone and the XY coordinates output by the filter is greater than the set value for multiple consecutive sampling cycles, the adaptive optimization of the Kalman gain K is started: The root mean square error is divided into three levels: low error, medium error and high error through the fuzzy logic controller, and the corresponding adjustment coefficient K value is assigned; when the error level jumps from low and medium error to high error, the strong tracking factor t is triggered to perform weighted correction on the state covariance matrix P of the Kalman filter.

8. The method according to claim 2, characterized in that When superimposed tension fluctuation suppression factors are added, they include: The high-frequency fluctuation component of the mooring line tension is collected and decomposed into three frequency bands through wavelet transform; Adaptive threshold noise reduction is performed on each frequency band component to retain the effective frequency band that represents the elastic vibration of the cable; The noise-reduced fluctuation component is converted into an inverse compensation PWM signal and linearly superimposed with the Z-axis PWM signal; The amplitude of the reverse compensation PWM signal is positively correlated with the root mean square value of the tension fluctuation, and the phase lag is set to 90° to achieve active vibration reduction.

9. The method according to claim 2, characterized in that When performing XYZ three-axis follow-by-wire control of the drone, it includes: The XY-axis PWM signal and the Z-axis PWM signal are nonlinearly decoupled using the coupling coefficient matrix of the tethered line's deployment length and the drone's real-time speed. The coupling coefficient matrix is ​​dynamically updated based on the tethered line's elastic coefficient model. An inverse UAV dynamics model is introduced to predict the UAV position deviation based on the current three-dimensional control signal. A feedforward compensation is generated and superimposed on the control signal. The compensation amplitude is positively correlated with the first-order derivative of the tether line tension fluctuation. When the tether line swing frequency exceeds the set value, the XY axis control weight is reduced and the attitude angle limit protection is activated. At the same time, the Z axis control is switched to constant tension priority mode, and the tension fluctuation is controlled through the PID parameter self-tuning algorithm.

10. A dual-mode dynamic cooperative control system based on pulling a mooring line, characterized in that: During operation, the method according to claim 1 is executed, including: Tethered ground equipment: includes a rocker arm pulley assembly, a tethering line, a tethering box winch, and a PWM signal generator. The rocker arm pulley assembly is equipped with a dual potentiometer for collecting displacement electrical signals. The PWM signal generator generates a Z-axis PWM signal based on the displacement electrical signal and divides it into two channels: one for transmission to the winch ESC and the other for transmission to the UAV. Drone: Includes flight control system, XY joystick, and signal merging module. The XY joystick generates XY-axis PWM signals based on the angle difference between the joystick and the tethered line. The signal merging module receives the Z-axis PWM signal sent by the tethered ground device and merges it with the XY-axis PWM signal to form a three-dimensional control signal. Switching execution unit: includes the first and second remote control relays and the remote control; the first and second remote control relays are each equipped with two sets of linkage contacts; the first remote control relay is used to control the control signal received by the flight control system, and the second remote control relay is used to control the control signal received by the winch electronic controller; the remote control is used to send the remote control signal of the drone and to control the conduction or disconnection of the remote control relay to make the drone enter the first or second control mode.

Images