A precise docking guidance method for a split flying car

Abstract

This invention relates to the field of flying car navigation and docking technology, and discloses a precise docking guidance method for a split-type flying car, comprising: acquiring rotor speed signals, extracting modulation components, tracking their instantaneous phase, determining obstruction phase, and adjusting mass weights; predicting signal interruption intervals, locking integer ambiguity, and maintaining ambiguity; when the modulation depth exceeds a threshold, fusing inertial navigation and visual pose, and performing phase alignment training; mapping the ratio of instantaneous phase to rotor speed to blade angle position, and increasing the safety domain radius when a pre-set danger angle window is available at this angle position; when both the sliding window standard deviation and random component energy exceed the threshold, determining an antenna mutation and suspending docking. This invention utilizes the periodic obstruction pattern of GNSS signals caused by rotor rotation, transforming passive response into active prediction, and achieving all-weather, highly robust centimeter-level precise docking.

Description

Technical Field

[0001] This invention relates to the field of flying car navigation and docking technology, and in particular to a precise docking guidance method for a split-type flying car. Background Technology

[0002] The split-type flying car consists of a flight module and a ground driving module, achieving flight and ground driving through mode switching. The key to mode switching lies in the flight module's precise landing on the driving module, achieving reliable docking. Existing docking guidance methods have the following technical shortcomings: First, pure GPS / RTK positioning solutions are prone to signal loss in obstructed environments such as urban canyons and under overpasses, leading to positioning interruptions and failing to meet the continuous high-precision requirements for docking.

[0003] Secondly, pure vision guidance solutions are greatly affected by environmental factors such as lighting, weather, and rotor airflow, resulting in poor robustness and limited effective recognition distance, making it difficult to cover the entire process from long distance to close distance.

[0004] Furthermore, existing integrated navigation (GNSS / INS) solutions suffer from rapid divergence of inertial navigation errors when GNSS signals are lost for extended periods. Lacking an effective intelligent compensation mechanism, they are unable to maintain positioning accuracy in scenarios where rotors periodically block signals.

[0005] Finally, the existing solution is not specifically designed to address the periodic blocking characteristics of GNSS signals caused by the rotor rotation of the flying car, and therefore cannot utilize this periodicity to optimize the docking guidance process.

[0006] Therefore, this invention proposes a precise docking guidance method for a split-type flying car. Summary of the Invention

[0007] The purpose of this invention is to solve the problems in the prior art by proposing a precise docking guidance method for a split-type flying car.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a precise docking guidance method for a split-type flying car, comprising the following steps: Step S1: Acquire rotor speed signal, extract modulation component with the same frequency as rotor speed from GNSS carrier-to-noise ratio sequence, track its instantaneous phase through phase-locked loop, determine the obstruction phase according to modulation depth, and dynamically adjust the quality weight of carrier phase observation value. Step S2: Based on the occlusion phase, predict the signal interruption interval, lock the integer ambiguity before the interruption arrives, keep the ambiguity unchanged within the interruption interval, and only update the position parameters; Step S3: When the modulation depth exceeds the threshold for multiple consecutive cycles, the inertial navigation and visual pose are fused through an extended Kalman filter, and the REF neural network is trained with instantaneous phase to compensate for periodic errors. Step S4: At the end of the docking phase, the ratio of the instantaneous phase to the rotor speed is mapped to the blade angle position. When the angle position falls into the preset danger angle window multiple times in a row, the safety zone radius is increased. Step S5: When the sliding window standard deviation of the instantaneous phase and the random component energy both exceed the threshold, it is determined that the antenna change is caused by airflow, the docking is suspended and step S1 is re-executed.

[0009] The beneficial effects of the technical solution provided by this invention include at least the following: This invention acquires rotor speed signals and extracts co-frequency modulation components from GNSS carrier-to-noise ratio sequences. By tracking the instantaneous phase through a phase-locked loop to determine the obstruction phase and adjusting the weight of carrier phase observations, it can actively identify rotor obstruction patterns and improve the accuracy of signal quality assessment. This invention predicts the signal interruption interval based on the occlusion phase and locks the integer ambiguity before the interruption arrives. Within the interruption interval, the ambiguity remains unchanged and only the position parameters are updated. This can change the passive response to the active prediction and avoid the loss of ambiguity lock during the interruption. This invention fuses inertial navigation and visual pose through an extended Kalman filter when the modulation depth exceeds a threshold for multiple consecutive cycles, and uses the instantaneous phase to train the REF neural network for phase alignment to compensate for periodic errors. This allows the neural network to predict errors and suppress inertial navigation divergence when GNSS signals are blocked. This invention maps the ratio of instantaneous phase to rotor speed to blade angular position at the end of the docking phase. When the angular position falls into the preset danger angle window multiple times in a row, the safety zone radius is increased, which can avoid the collision risk caused by the rotor blade sweeping across the antenna's field of view in advance. This invention identifies non-periodic airflow interference and forms a closed-loop adaptive mechanism by determining that the antenna change is caused by airflow when the sliding window standard deviation of the instantaneous phase and the random component energy both exceed the threshold, pausing the docking and re-executing step S1. Attached Figure Description

[0010] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention; Figure 2 The flowchart for interruption prediction and ambiguity pre-locking provided in the embodiments of the present invention. Detailed Implementation

[0012] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a precise docking guidance method for a split-type flying car proposed according to the present invention. In the following description, different embodiments or one embodiment do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0014] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0015] The following description, in conjunction with the accompanying drawings, details a specific scheme for a precise docking guidance method for a split-type flying car provided by the present invention.

[0016] Please see Figure 1 The diagram illustrates a method flow chart of a precise docking guidance method for a split-type flying car according to an embodiment of the present invention, including the following steps: Step S1: Acquire rotor speed signal, extract modulation component with the same frequency as rotor speed from GNSS carrier-to-noise ratio sequence, track its instantaneous phase through phase-locked loop, determine the obstruction phase according to modulation depth, and dynamically adjust the quality weight of carrier phase observation value. Step S2: Based on the occlusion phase, predict the signal interruption interval, lock the integer ambiguity before the interruption arrives, keep the ambiguity unchanged within the interruption interval, and only update the position parameters; Step S3: When the modulation depth exceeds the threshold for multiple consecutive cycles, the inertial navigation and visual pose are fused through an extended Kalman filter, and the REF neural network is trained with instantaneous phase to compensate for periodic errors. Step S4: At the end of the docking phase, the ratio of the instantaneous phase to the rotor speed is mapped to the blade angle position. When the angle position falls into the preset danger angle window multiple times in a row, the safety zone radius is increased. Step S5: When the sliding window standard deviation of the instantaneous phase and the random component energy both exceed the threshold, it is determined that the antenna change is caused by airflow, the docking is suspended and step S1 is re-executed.

[0017] It should be noted that the rotor speed signal refers to the angular velocity or frequency information of the rotor rotation of the flight module, in revolutions per minute or Hertz, and can be obtained through Hall sensors, photoelectric encoders or from the rotor motor control signals.

[0018] GNSS carrier-to-noise ratio sequence refers to a data sequence formed by arranging the carrier-to-noise ratio values ​​output by a global navigation satellite system receiver in chronological order. The unit is decibel-hertz (dBHZ), and it reflects the quality of the received signal.

[0019] The modulation component refers to the periodic fluctuation component in the GNSS carrier-to-noise ratio sequence that is the same as or an integer multiple of the rotor rotation frequency. This fluctuation is caused by the rotor blades periodically blocking the GNSS antenna.

[0020] A phase-locked loop (PLL) is a closed-loop control system that can track changes in the phase of an input signal. Its output can lock the instantaneous phase of the input signal, maintaining tracking even when the signal fluctuates.

[0021] Instantaneous phase refers to the phase value output by the phase-locked loop that is synchronized with the input modulation component, reflecting the angular position of the rotor blade relative to the GNSS antenna at any given moment.

[0022] Modulation depth refers to the magnitude of periodic fluctuations in the GNSS carrier-to-noise ratio sequence, specifically the ratio of peak-to-peak carrier-to-noise ratio to mean carrier-to-noise ratio, used to quantify the degree of obstruction of GNSS signals by the rotor.

[0023] The blocking phase refers to the instantaneous phase value corresponding to the direction when the rotor blade rotates to the line connecting the GNSS antenna and the satellite. Within this phase range, the GNSS signal is blocked.

[0024] Carrier phase observations refer to the carrier phase measurements output by a GNSS receiver, measured in cycles, and are used for high-precision positioning calculations.

[0025] Quality weight refers to the confidence coefficient assigned to each observation in the positioning solution. The higher the weight, the greater the contribution of that observation in the solution.

[0026] Integer ambiguity refers to the uncertainty of integer multiple wavelengths in GNSS carrier phase observations, and its fixation is the key to centimeter-level positioning.

[0027] The interruption interval refers to the period during which the rotor blades block the GNSS antenna, causing a decrease in signal quality. During this interval, carrier phase observations are unavailable or unreliable.

[0028] The Extended Kalman Filter (EPF) is a recursive filtering algorithm used for state estimation of nonlinear systems. It can fuse data from multiple sensors and output the optimal estimate.

[0029] REF neural network refers to radial basis function neural network, which is a three-layer neural network model that uses Gaussian function as activation function and is used for nonlinear function approximation.

[0030] Phase alignment training refers to organizing the training samples of a neural network according to the temporal sequence of instantaneous phases, so that the network learns the error characteristics related to the rotor rotation phase.

[0031] The blade angular position refers to the angle value obtained by mapping the ratio of instantaneous phase to rotor speed to a range of 0 to 360 degrees, representing the spatial orientation of the rotor blade at the current moment.

[0032] The danger angle window refers to a preset angle range. When the blade angle falls into this range, it means that the rotor blade is about to enter the GNSS antenna's field of view, posing a risk of signal blockage.

[0033] The safety zone radius refers to the radius of the circular area centered on the ground module docking platform, which allows the flight module projection point to move freely. It is used to determine the safety of docking.

[0034] The sliding window standard deviation is the standard deviation calculated for an instantaneous phase sequence within a fixed-length phase sliding window, used to quantify the degree of phase fluctuation.

[0035] Random component energy refers to the percentage of energy extracted from the phase sequence after performing a Fourier transform on the spectral components that are independent of the rotor speed fundamental frequency and its harmonics, and is used to identify non-periodic interference.

[0036] Antenna mutation refers to the phenomenon where the phase center of a GNSS antenna shifts instantaneously due to rotor airflow disturbance, manifesting as an abnormal jump in the instantaneous phase.

[0037] As a specific implementation method, a certain type of quadcopter split-type flying car will be used as an example. The flying car consists of a flight module and a ground driving module. The flight module has a wheelbase of 2.5 meters and is equipped with a Huace X1 RTK board, an AVIC INS300 inertial navigation system, a Jetson TX2 airborne computer, and a downward-looking USB camera.

[0038] A composite visual identifier, combining AprilTag and ArUco design, is deployed at the top center of the ground module. The outer identifier measures 15cm x 15cm and has an effective recognition distance of 5m to 1m. The middle identifier measures 5cm x 5cm and has an effective recognition distance of 1m to 0.3m. An inner QR code array is used for precise positioning within 0.3m. A monocular camera with a resolution of 1280720 pixels and a frame rate of 30 frames per second is installed at the bottom of the flight module.

[0039] After takeoff, the flight module obtains centimeter-level absolute coordinates from the base station network based on real-time dynamic differential positioning technology and broadcasts them to the ground module. When the flight module descends to a distance of 50 meters from the ground module, the docking guidance process of this invention is initiated.

[0040] Step S1 is implemented as follows: Accelerometers (model PCB352C33) are installed at the root of each of the four rotor blades on the flight module as vibration sensors to collect the flapping vibration signals of the blades at a sampling frequency of 2000Hz. A Fourier transform is performed on the collected flapping vibration signals to obtain a spectrum. In the spectrum, the fundamental frequency component appears at 20Hz, corresponding to the fundamental frequency when the rotor rotates at 1200 revolutions per minute. This fundamental frequency component at 20Hz is extracted as the rotor speed signal.

[0041] Sideband components appear on both sides of the fundamental frequency component, with the amplitude of the 20.5Hz sideband on the right being 12% of the fundamental frequency amplitude. Calculating the amplitude ratio of the sideband component to the fundamental frequency component yields an imbalance of 0.12. The preset imbalance threshold is 0.1; the current imbalance of 0.12 exceeds the threshold, indicating a slight mechanical imbalance in the rotor. At this point, the loop bandwidth of the phase-locked loop is temporarily increased from the normal operating 2Hz to 1.5 times the fundamental frequency of 20Hz, i.e., 30Hz.

[0042] An adaptive bandpass filter is applied to the GNSS carrier-to-noise ratio sequence. The filter's center frequency tracks the rotor speed fundamental frequency of 20Hz in real time, and the bandwidth is set to 2Hz. A frequency traction auxiliary loop is introduced, with the target range for frequency traction set to 10% of the rotor speed fundamental frequency of 20Hz, i.e., 18Hz to 22Hz. When the instantaneous phase error of the phase-locked loop exceeds 30 degrees, the center frequency is pulled back to the 18Hz to 22Hz range through the frequency traction auxiliary loop.

[0043] By utilizing the phase difference output from the phase-locked loop phase detector and combining it with the rotor blade flapping phase extracted from the flapping vibration signal, the dynamic deformation of the rotor blade in the plane of rotation is calculated. The calculated dynamic deformation is 3% of the blade length. A preset proportional threshold of 2% is set; if the dynamic deformation exceeds 3%, it is determined that the rotor has undergone significant elastic deformation.

[0044] The correction is calculated based on the 3% dynamic deformation of the rotor and the blade length. The preset angle coefficient is 90 degrees, and the correction is 0.03 multiplied by 90 degrees, which equals 2.7 degrees. The theoretical spatial angle between the rotor blade and the GNSS antenna is 35 degrees, and the corrected spatial angle is 32.3 degrees.

[0045] The modulation depth (MDD) is calculated based on the filtered GNSS carrier-to-noise ratio (CNR) sequence, and is the ratio of the peak-to-peak CNR to the mean CNR. 100 sampling points are collected within one rotor cycle at a sampling frequency of 2000 Hz. The number of sampling points with a MDD exceeding 3 dB within a single rotor cycle is 78, accounting for 78%, exceeding the 70% threshold, and is thus determined to be the start of a phase obstruction. In two consecutive rotor cycles, this percentage is 28% and 25% respectively, both below 30%, and is thus determined to be the end of a phase obstruction.

[0046] The real-time angular distance between the rotor blades and the GNSS antenna is calculated based on the instantaneous phase output from the phase-locked loop. When the angular distance is 28 degrees, which is less than the 30-degree threshold, pre-weighting is initiated, reducing the quality weight of the carrier phase observation from the original weight of 1.0 to 50% of the target weight of 0.5, i.e., 0.25. When the angular distance is 8 degrees, which is less than the 10-degree threshold, the quality weight is reduced to the target weight of 0.5.

[0047] Within the unobstructed phase interval, the instantaneous phase of the phase-locked loop (PLL) is smoothed using carrier phase observations, employing a mean filter with a sliding window length of 10 sampling points. The smoothed phase is used to correct the phase prediction value within the obstructed interval of the next cycle. The deviation between the smoothed phase and the predicted phase in the unobstructed interval of the current cycle is 1.5 degrees. This deviation is added as a correction factor to the phase prediction of the obstructed interval of the next cycle, forming a closed-loop correction for phase prediction.

[0048] The specific implementation of step S2 is as follows: Based on the instantaneous phase output by the phase-locked loop and the rotor speed, combined with the number of blades of the four-bladed rotor, the predicted blocking phase at each moment in the next two rotor cycles is calculated. For a four-bladed rotor, there are 4 equally spaced blocking windows within each rotor cycle, with a window width of 45 degrees.

[0049] Calculate the prediction error covariance. Assume the instantaneous standard deviation of rotor speed is 2 revolutions per minute, which translates to a frequency standard deviation of 0.0333 Hz. The prediction error covariance equals the square of the instantaneous standard deviation of rotor speed multiplied by a preset scaling factor of 0.5. That is, the square of 0.0333 is 0.00111, multiplied by 0.5, equals 0.000555 Hz squared. The prediction error standard deviation is 0.0236 Hz, which translates to an angle of 0.425 degrees.

[0050] The rotor rotation state is determined based on the deviation between the predicted obstruction phase and the measured obstruction phase of the previous cycle. When the absolute value of the deviation is less than 0.5 times the standard deviation of the prediction error (i.e., 0.2125 degrees) for three consecutive times, it is determined to be in a steady state, and the interruption prediction mode is activated. When it is determined to be in a transient state, the interruption prediction is skipped, and the system switches to pure inertial navigation mode, which continues to run until the instantaneous standard deviation of the rotor speed recovers to below 1 revolution per minute, at which point the execution restarts from step S2-1.

[0051] Before the interruption begins, the prediction error covariance is normalized to the [0,1] interval. Assuming the maximum covariance is 0.002, the current covariance is 0.000555, and the normalized value C is 0.2775. The elevation angle threshold is calculated using the formula T = 15 + 30C, where T = 15 + 300.2775 = 23.3 degrees, rounded to 23 degrees. Only satellites with elevation angles greater than 23 degrees participate in ambiguity fixing.

[0052] The fixed integer ambiguity values ​​are stored in a non-volatile register and frozen during the interrupt interval. Within the interrupt interval, the position is updated by accumulating the velocity increments recursively derived from inertial navigation, using the carrier phase position solution at the start of the interrupt as the initial value, while maintaining the integer ambiguity unchanged. The interrupt interval lasts for 6.25 milliseconds, with a velocity increment of 0.003125 meters. The updated positions are 100.003125 meters east, 200.001875 meters north, and 50.000625 meters upwards.

[0053] After the interruption period ends, the difference between the initial carrier phase value recorded at the start of the interruption and the final carrier phase value recorded at the end of the interruption is calculated to inversely determine the integer ambiguity change. Assuming the initial carrier phase value is 16.3 cycles and the final value is 18.9 cycles, the difference is 2.6 cycles. The theoretical change is 0.125 cycles, and the inversely calculated absolute change is 2.475 cycles. The preset change threshold is 0.3 cycles; if 2.475 cycles exceeds the threshold, a cycle slip is determined, triggering ambiguity re-fixing.

[0054] During refixation, the last integer ambiguity before the interruption is used as prior information to convert into pseudo-observations. The position change of 0.003125 meters recursively calculated within the interruption interval is used as a geometric constraint. The LAMBDA algorithm is executed, and only the subset of satellites with an absolute value of back-calculated change exceeding 0.3 weeks is refixed.

[0055] The specific implementation of step S3: When the modulation depth exceeds 6dB for five consecutive rotor cycles, the GNSS signal is determined to be blocked. Based on the instantaneous phase output of the phase-locked loop, each rotor cycle is divided into four equally spaced phase intervals, each interval being 90 degrees wide. An independent REF neural network subnetwork is established for each phase interval.

[0056] During normal periods when the modulation depth is less than 6 dB, the corresponding sub-network is activated based on the current instantaneous phase for training, and only the weights of that sub-network are updated. When the instantaneous phase is 45 degrees, the first sub-network is activated, and the position residual of 0.05 meters is used as the supervision signal. The weights are updated using the stochastic gradient descent method with momentum, a momentum factor of 0.9, and a learning rate of 0.001.

[0057] Under occlusion conditions, the corresponding sub-network is activated based on the current instantaneous phase to predict errors. Assuming the instantaneous phase is 135 degrees, the second sub-network is activated, outputting predicted errors of 0.08 meters eastward, 0.05 meters northward, and 0.12 meters upward. Zero-bias correction is then performed by subtracting the historical averages of 0.03 meters eastward, 0.02 meters northward, and 0.04 meters upward, resulting in corrected values ​​of 0.05 meters eastward, 0.03 meters northward, and 0.08 meters upward.

[0058] The corrected error predictions and the inertial navigation recursive results are fused using an extended Kalman filter with a filtering period of 10 milliseconds. The observation noise covariance is dynamically adjusted based on the historical prediction accuracy of the current phase interval, with a setting of 0.04 for the second interval (higher accuracy) and 0.16 for the fourth interval (lower accuracy).

[0059] After the occlusion ends, the subnetworks with prediction accuracy below 0.05 meters are quickly corrected. The average prediction error of the third subnetwork is 0.12 meters, and that of the fourth subnetwork is 0.09 meters. The learning rate is increased to 0.003, and after 50 iterations, the errors are reduced to 0.045 meters and 0.042 meters, respectively. The learning rate is then restored to 0.001.

[0060] The specific implementation of step S4: When the distance between the docking end section and the ground module is less than 5 meters, the instantaneous phase output of the phase-locked loop is mapped to the blade angle position from 0 to 360 degrees. The phase-locked loop output is 1.57 radians, which is 90 degrees, and the blade angle position is 90 degrees.

[0061] The uncertainty of the angular position is calculated based on the root mean square of the phase error over the past rotor cycle using the phase-locked loop. The root mean square of the 50 phase error values ​​is 1.4 degrees, and the uncertainty is 1.4 degrees.

[0062] According to the preset danger angle window of the four-bladed rotor, the window width is 45 degrees, and the center of the window is located at 0 degrees where the direction of the line connecting the blade and the GNSS antenna coincides, with a range of -22.5 degrees to 22.5 degrees, or 337.5 degrees to 22.5 degrees.

[0063] The difference between the angular position and its uncertainty is used as the lower confidence limit. When the blade angular position is 10 degrees, the lower confidence limit is 8.6 degrees, falling into the danger angle window, and the first record is made. When the blade angular position is 9 degrees, the lower confidence limit is 7.6 degrees, falling into the danger angle window again, and the second record is made. When the blade angular position is 8 degrees, the lower confidence limit is 6.6 degrees, falling into the danger angle window again, and the third record is made. After three consecutive occurrences of this, it is determined that the rotor blade is about to sweep across the GNSS antenna's field of view.

[0064] Based on the flight module's current altitude of 5 meters and descent speed of 0.5 meters per second, the basic risk coefficient is calculated to be 0.1. Multiplying this coefficient by the root mean square of the phase error of 1.4 degrees yields a corrected risk coefficient of 0.14. The safety zone radius is then increased to 1.14 times the original radius of 0.3 meters, i.e., 0.342 meters.

[0065] The specific implementation of step S5 is as follows: Real-time monitoring of the instantaneous phase output of the phase-locked loop, constructing a phase sliding window with a length of 10 rotor cycles, a sampling interval of 1 millisecond, and the window containing 500 phase data points.

[0066] Calculate the standard deviation of the phase within the window. Under normal conditions, the standard deviation is 2 degrees, below the 5-degree threshold. When airflow disturbances occur, the standard deviation rises to 8 degrees, exceeding 5 degrees, thus satisfying the standard deviation condition.

[0067] A Fourier transform is performed on the phase sequence within the window to extract random components in the spectrum that are independent of the rotor speed fundamental frequency (20Hz) and its harmonics. Assuming a total energy of 50000, a fundamental frequency and harmonic energy of 15700, and a random component energy of 34300 (68.6%), exceeding the 40% threshold, the random component condition is satisfied.

[0068] When the standard deviation condition (8 degrees > 5 degrees) and the random component condition (68.6% > 40%) are met simultaneously, it is determined that the antenna phase center change is caused by the rotor airflow, and the instantaneous phase of 135 degrees at the moment of change is recorded as the interference phase.

[0069] A pause docking command is generated to stop the descent motion of the flight module and re-execute step S1. Upon re-execution, the recorded interference phase of 135 degrees is used as the initial phase estimate of the phase-locked loop (PLL). The PLL starts tracking from around 135 degrees, saving 135 degrees of lock time, or 18.75 milliseconds, and reducing the relock time by 37.5%.

[0070] In the above embodiments, the flight module uses RTK carrier phase differential positioning within a distance of 50 meters to 5 meters from the ground module, with the horizontal positioning error controlled within 2 centimeters. Within a distance of 5 meters to 0.3 meters, visual marker-assisted positioning is used, with the horizontal deviation controlled within 1 centimeter. Within 0.3 meters, final docking is completed using the geometric constraints of the guidance mechanism. Throughout the docking process, the safety domain radius is dynamically adjusted based on real-time risks, and airflow interference can be promptly identified and a retry mechanism triggered, ensuring the reliability and safety of the docking.

[0071] Step S1 further includes the following sub-steps: S1-1, A vibration sensor is installed at the root of the rotor blade. The collected flapping vibration signal is subjected to Fourier transform, the fundamental frequency component is extracted as the rotor speed signal, and the rotor imbalance is judged from the sideband amplitude of the fundamental frequency component. S1-2, the rotor imbalance is determined by the amplitude ratio of the sideband component to the fundamental frequency component. When the imbalance exceeds the preset imbalance threshold, the loop bandwidth of the phase-locked loop is temporarily increased to a preset multiple of the fundamental frequency to suppress sideband interference. S1-3, adaptive filtering of GNSS carrier-to-noise ratio sequence, setting the center frequency of the filter to the rotor speed base frequency, and introducing a frequency traction auxiliary loop. When the instantaneous phase error of the phase-locked loop exceeds the preset angle threshold, the center frequency is pulled back to the preset frequency range of the rotor speed base frequency through the frequency traction auxiliary loop. S1-4: Using the phase difference output by the phase-locked loop phase detector and the flapping phase of the rotor blades, calculate the dynamic deformation of the rotor blades in the plane of rotation. When the dynamic deformation exceeds the preset proportional threshold of the blade length, it is determined that the rotor has generated elastic deformation. S1-5, calculates the correction amount based on the magnitude of the rotor dynamic deformation and the blade length, corrects the spatial angle between the rotor blade and the GNSS antenna line, and the correction amount is the ratio of deformation to blade length multiplied by a preset angle coefficient. S1-6, calculate the modulation depth based on the filtered GNSS carrier-to-noise ratio sequence, determine the obstruction phase based on the magnitude of the modulation depth, and adjust the quality weight of the carrier phase observation value in stages. The modulation depth is the ratio of the peak-to-peak value of the carrier-to-noise ratio to the mean value of the carrier-to-noise ratio. S1-7, within the unobstructed phase interval, the instantaneous phase of the phase-locked loop is smoothed using the carrier phase observation value. The smoothed phase is used to correct the phase prediction value in the obstructed interval of the next cycle, forming a closed-loop correction of the phase prediction.

[0072] Furthermore, in sub-steps S1-6, determining the blocking phase based on the modulation depth includes: determining the number of sampling points in a single rotor cycle based on the rotor speed; using the proportion of sampling points with a modulation depth exceeding 3dB in a single rotor cycle exceeding 70% as the starting condition for the blocking phase; and using the proportion of this proportion being less than 30% in two consecutive rotor cycles as the ending condition for the blocking phase.

[0073] Furthermore, in sub-steps S1-6, the graded adjustment includes: calculating the real-time angular distance between the rotor blade and the GNSS antenna based on the instantaneous phase output by the phase-locked loop; initiating pre-weight reduction when the angular distance is less than 30 degrees, reducing the quality weight of the carrier phase observation to 50% of the target weight; and reducing it to the target weight when the angular distance is less than 10 degrees, thus forming a continuous two-level weight reduction mechanism related to the angular distance.

[0074] It should be noted that vibration sensors refer to accelerometers or piezoelectric sensors installed at the root of rotor blades to collect flapping vibration signals generated when the blades rotate.

[0075] The flapping vibration signal refers to the periodic vibration waveform generated by the rotor blades in the plane of rotation. Its frequency is related to the rotor speed, and its amplitude reflects the dynamic balance state of the blades.

[0076] Fourier transform is a mathematical transformation method that converts a time-domain vibration signal into a frequency-domain spectrum, used to extract the fundamental frequency component and sideband components of the signal.

[0077] The fundamental frequency component is the frequency component with the largest amplitude in the vibration signal spectrum of the rotor. This frequency is equal to the rotor speed multiplied by the number of blades, representing the rotor's rotational frequency.

[0078] Sideband components refer to the frequency components distributed on both sides of the fundamental frequency in the spectrum. They are caused by rotor imbalance or elastic deformation, and their amplitude reflects the mechanical state of the rotor.

[0079] Imbalance is the ratio of the amplitude of the sideband component to the amplitude of the fundamental frequency component. It is used to quantify the dynamic balance state of the rotor. The larger the value, the more severe the imbalance.

[0080] Loop bandwidth refers to the frequency range of the phase-locked loop's response. The wider the bandwidth, the faster the tracking speed but the weaker the noise immunity; the narrower the bandwidth, the smoother the tracking but the slower the response.

[0081] Sideband interference refers to the negative impact of non-fundamental frequency components caused by rotor imbalance on the tracking performance of phase-locked loops, which may lead to phase-locking instability.

[0082] Adaptive filtering is a filtering method in which the center frequency of the filter is adjusted in real time according to the rotor speed. It is used to extract the modulation component with the same frequency as the rotor speed from the GNSS carrier-to-noise ratio sequence.

[0083] The center frequency refers to the center position of the passband of the bandpass filter. When set to the rotor speed fundamental frequency, it can extract the same frequency modulation component from the GNSS carrier-to-noise ratio sequence and filter out other frequency components.

[0084] Frequency traction auxiliary loop refers to a frequency acquisition mechanism that assists the operation of a phase-locked loop (PLL). When the PLL loses lock, it forcibly pulls the center frequency back to the vicinity of the target frequency.

[0085] Instantaneous phase error refers to the difference between the instantaneous phase of the phase-locked loop output and the true phase of the input signal, and is used to determine whether the phase-locked loop is in a locked state.

[0086] A phase detector is a circuit module in a phase-locked loop used to compare the phase difference between the input signal and the output signal. The output phase difference signal is used for loop filtering and voltage-controlled oscillator adjustment.

[0087] The flapping phase refers to the phase information extracted from the flapping vibration signal, which reflects the instantaneous angular position of the rotor blades during the rotation cycle.

[0088] Dynamic deformation refers to the elastic deformation amplitude of rotor blades under the action of centrifugal force and aerodynamic force, which is manifested as the displacement of the blades from the ideal rigid position.

[0089] Elastic deformation refers to the deformation of a rotor blade due to force during rotation. When the amount of deformation exceeds a preset proportional threshold of the blade length, it is judged as significant elastic deformation.

[0090] The spatial angle refers to the geometric angle between the rotor blade and the line connecting the GNSS antenna, and is used to determine whether the rotor blade is in a position that blocks the antenna.

[0091] The correction amount refers to the compensation value that adjusts the theoretical spatial angle based on the dynamic deformation amount. It is equal to the ratio of the deformation amount to the blade length multiplied by the preset angle coefficient.

[0092] Graded adjustment refers to a method of adjusting the quality weight by dividing it into multiple levels according to the degree of occlusion. The weight is reduced more significantly when there is full occlusion and less significantly when there is partial occlusion.

[0093] Pre-weight reduction refers to the operation of reducing the mass weight in advance before the rotor blades enter the obstruction zone, in order to smooth the transition and avoid abrupt changes in weight.

[0094] The target weight refers to the final value that the quality weight should reach when occlusion occurs completely, and it serves as the target for weight reduction within the full occlusion range.

[0095] The unobstructed phase interval refers to the instantaneous phase interval of the GNSS antenna when the rotor blades do not obstruct it, and the carrier phase observations within this interval are of good quality.

[0096] Smoothing filtering refers to a method of filtering the instantaneous phase of a phase-locked loop using high-quality observations to reduce phase noise and jitter.

[0097] Closed-loop correction refers to a feedback mechanism that uses the filtered phase of the unobstructed region to correct the phase prediction value of the obstructed region, forming an adaptive phase prediction closed loop.

[0098] The number of sampling points refers to the number of times the modulation depth is sampled within one rotor cycle, which is determined by both the rotor speed and the sampling frequency.

[0099] Angular distance refers to the real-time spatial angle difference between the rotor blades and the GNSS antenna, which is calculated from the instantaneous phase output by the phase-locked loop.

[0100] The two-stage weight reduction mechanism refers to a strategy that reduces the quality weight in two stages based on the size of the angular distance. When the angular distance is large, the weight is pre-reduced, and when the angular distance is small, the weight is reduced to the target weight.

[0101] Step S1 of this application overcomes the limitation of existing technologies that cannot actively identify the obstruction phase using the periodicity of the rotor by performing cross-domain coupling analysis between the rotor mechanical vibration signal and the GNSS carrier-to-noise ratio sequence. Specifically, a vibration sensor is installed at the root of the rotor blade to collect the flapping vibration signal, and the fundamental frequency is extracted as the rotor speed signal through Fourier transform. At the same time, the rotor imbalance is determined from the sideband components. This step is equivalent to obtaining the precise rotational speed and mechanical state of the rotor using mechanical vibration information. Adaptive filtering is performed on the GNSS carrier-to-noise ratio sequence to extract the modulation component with the same frequency as the rotor rotation, and a frequency traction auxiliary loop is introduced to avoid the phase-locked loop locking to harmonics. Since the rotor rotation has strict periodicity, its obstruction of the GNSS antenna is directly related to the instantaneous phase. By tracking the instantaneous phase through the phase-locked loop and combining it with dynamic deformation to correct the spatial angle, the obstruction phase can be accurately determined. Using a graded weighting mechanism driven by modulation depth ratio and angular distance, the weight of the carrier phase observation value can be smoothly adjusted according to the degree of obstruction. Closed-loop correction in the non-obstruction interval continuously improves the phase prediction accuracy over time. The entire process forms a cross-domain fusion chain from mechanical vibration analysis to GNSS signal processing, enabling the split-type flying car to actively predict signal blockage based on the mechanical characteristics of the rotor, rather than passively responding to signal loss.

[0102] Please see Figure 2 The flowchart for interruption prediction and ambiguity pre-locking provided in the embodiments of the present invention.

[0103] Step S2 further includes the following sub-steps: S2-1, based on the instantaneous phase output of the phase-locked loop and the rotor speed, combined with the number of rotor blades, calculate the predicted blocking phase at each moment in the next two rotor cycles, and calculate the prediction error covariance of the blocking phase. The prediction error covariance is the square of the instantaneous standard deviation of the rotor speed multiplied by a preset scaling factor. S2-2: Based on the deviation between the predicted blocking phase and the measured blocking phase of the previous cycle, determine the rotor rotation state. If it is determined to be in a steady state, execute S2-3 to S2-6. If it is determined to be in a transient state, skip S2-3 to S2-6, switch to pure inertial navigation mode, and continue to run pure inertial navigation mode until the instantaneous standard deviation of rotor speed recovers to below the preset standard deviation threshold, and start executing again from S2-1. S2-3, before the interruption start time arrives, normalize the prediction error covariance to the [0,1] interval, and adjust the elevation angle threshold in the ambiguity fixing strategy according to the magnitude of the prediction error covariance, so that the threshold changes linearly between 15 degrees and 45 degrees. The specific formula is T=15+30C, where C is the normalized covariance value. S2-4: Store the fixed integer ambiguity value into a non-volatile register and freeze the register during the interrupt interval; S2-5, within the interruption interval, using the carrier phase position solution at the start of the interruption as the initial value, the position is updated by using the velocity increment recursively derived by inertial navigation, keeping the integer ambiguity unchanged; S2-6 After the interruption interval ends, the difference between the initial carrier phase value recorded at the start time of the interruption interval and the final carrier phase value recorded at the end time of the interruption interval is used to calculate the integer ambiguity change in the interruption interval. When the absolute value of the back-calculated change exceeds the preset change threshold, it is determined that a cycle slip has occurred in the interruption interval, triggering ambiguity re-fixing.

[0104] Furthermore, in sub-step S2-2, the step of determining the rotor rotation state includes: Calculate the absolute value of the deviation between the predicted occlusion phase and the measured occlusion phase of the previous cycle, and obtain the standard deviation of the prediction error; When the absolute value of the deviation is less than 0.5 times the standard deviation of the prediction error for three consecutive times, it is determined that the rotor has entered a steady state. At this time, the predicted occlusion phase is reliable, and the interruption prediction mode is started. Otherwise, it is judged as transient, in which case the prediction is unreliable, and interruption of the prediction is prohibited.

[0105] Furthermore, in sub-steps S2-6, the steps to trigger ambiguity re-fixing include: The last integer ambiguity before the interruption interval is used as prior information, and it is converted into a pseudo-observation value and added to the double-difference observation equation to constrain the calculation of the floating-point ambiguity solution; By using the position change recursively calculated within the interruption interval as a geometric constraint, the ambiguity search space is reduced; The LAMBDA algorithm is used to fix the integer ambiguity, and only the subset of satellites whose absolute value of the back-calculated change exceeds the preset change threshold is re-fixed, while the integer ambiguity of the remaining satellites remains unchanged from the fixed value before the interruption.

[0106] It should be noted that rotor speed refers to the angular velocity or frequency of the rotor of the flight module, measured in Hertz or revolutions per minute.

[0107] The number of rotor blades refers to the number of blades on a single rotor of a flight module, commonly two, three, or four blades.

[0108] Predicted blocking phase refers to the instantaneous phase value of GNSS signal blocking that is expected to occur within the future rotor cycle, calculated based on the instantaneous phase output by the phase-locked loop and the rotor speed.

[0109] Prediction error covariance is a quantitative indicator of the uncertainty of the predicted occlusion phase. It is equal to the square of the instantaneous standard deviation of the rotor speed multiplied by a preset proportional coefficient and is used to evaluate the reliability of the prediction.

[0110] The instantaneous standard deviation of rotor speed refers to the statistical standard deviation of rotor speed fluctuations over a short period of time, reflecting the stability of rotor rotation.

[0111] The measured phase of obstruction refers to the instantaneous phase value obtained by actual measurement through a phase-locked loop when the rotor blades obstruct the GNSS antenna.

[0112] Steady state refers to the rotational state in which the rotor speed is stable and the deviation between the predicted blocking phase and the measured blocking phase is small. At this time, the prediction result is reliable.

[0113] Transient refers to a rotational state in which the rotor speed fluctuates greatly and the predicted blocking phase deviates significantly from the measured blocking phase. In this state, the prediction results are unreliable.

[0114] Pure inertial navigation mode refers to a working mode that relies solely on the inertial navigation system for position and attitude calculation, without depending on GNSS observations.

[0115] The interruption start time refers to the moment when the rotor blades begin to block the GNSS antenna, which is the starting point of the signal interruption interval.

[0116] The elevation angle threshold is a threshold value used to filter satellites. Only satellites with an elevation angle greater than this threshold can participate in ambiguity fixing.

[0117] Normalization is a method of mapping the prediction error covariance to the [0,1] interval, so that data of different dimensions are comparable.

[0118] Non-volatile registers are storage units that do not lose data after power failure, and are used to maintain integer ambiguity values ​​within interrupted intervals.

[0119] The carrier phase position solution refers to the position information calculated based on carrier phase observations, with an accuracy down to the centimeter level.

[0120] Inertial navigation recursion refers to the method of obtaining position, velocity, and attitude by integrating and recursively calculating the angular velocity and acceleration measured by gyroscopes and accelerometers.

[0121] Velocity increment refers to the change in velocity obtained by the inertial navigation system recursively per unit time, which is used for position accumulation and update.

[0122] Cycle slip refers to the phenomenon where integer ambiguity changes abruptly in carrier phase observations, which manifests as a disruption of the continuity of phase measurements.

[0123] Prior information refers to the integer ambiguity value known before the interruption interval, which serves as the initial reference when re-fixing.

[0124] Pseudo-observations refer to the use of prior information to construct virtual observation equations and add them to filters to constrain the calculation of floating-point solutions.

[0125] The double-difference observation equation is an observation model obtained by performing double-difference processing on the carrier phase observations of two satellites and two receivers, which can eliminate receiver and satellite clock errors.

[0126] Geometric constraints refer to the restrictions imposed on the ambiguity search space by the recursive position change amount of inertial navigation.

[0127] The LAMBDA algorithm refers to the least squares ambiguity decorrelation adjustment algorithm, which is used for efficient searching and fixing integer ambiguities.

[0128] A satellite subset refers to a set of satellites selected from all visible satellites, and only the satellites in this set are subject to ambiguity refixation.

[0129] The standard deviation of prediction error is the square root of the covariance of prediction error, used to quantify the degree of deviation between the predicted occlusion phase and the measured occlusion phase.

[0130] Step S2 of this application overcomes the limitations of existing technologies that cannot handle rotor speed fluctuations and cannot detect cycle slips during interruptions by introducing a steady-state discrimination mechanism based on prediction error covariance and a cycle slip back-calculation verification mechanism before and after the interruption. Specifically, the predicted obstruction phase and its error covariance for the next two rotor cycles are calculated based on the instantaneous phase output by the phase-locked loop and the rotor speed. The prediction error covariance reflects the stability of the rotor speed; the smaller the covariance, the smoother the rotor rotation and the more reliable the prediction result. When the rotor is in a steady state, the predicted obstruction phase is reliable, and interruption prediction and ambiguity pre-locking are performed; when the rotor is in a transient state, the prediction is unreliable, and the system switches to pure inertial navigation mode to avoid ambiguity loss due to erroneous prediction. Before the interruption starts, the elevation angle threshold is dynamically adjusted based on the prediction error covariance; the larger the covariance, the higher the threshold, and only the most reliable satellite is fixed. During the interruption, the position is updated using the velocity increment recursively derived from inertial navigation, keeping the ambiguity unchanged. After the interruption ends, the difference in carrier phase before and after the interruption is used to back-calculate whether a cycle slip occurred during the interruption. If a cycle slip occurs, the ambiguity before the interruption is used as prior information, and only the affected subset of satellites is re-fixed. The entire process forms an adaptive chain from steady-state discrimination and dynamic threshold adjustment to cycle slip back-calculation verification and partial re-fixing, enabling the split-type flying car to maintain positioning reliability when the rotor speed fluctuates.

[0131] Step S3 further includes the following sub-steps: S3-1, based on the instantaneous phase output of the phase-locked loop, divide each rotor cycle into K equally spaced phase intervals, and establish an independent REF neural network sub-network for each phase interval, where K is an integer multiple of the number of rotor blades; S3-2, when the modulation depth does not exceed the modulation depth threshold, activate the corresponding sub-network according to the current instantaneous phase for training and update the weights of the sub-network; S3-3, Set the upper limit of continuous cycle count. When the number of cycles in which the modulation depth continuously exceeds the modulation depth threshold reaches or exceeds the upper limit, it is determined that the GNSS signal is in a blocked state. S3-4, In the occlusion state, activate the corresponding sub-network according to the current instantaneous phase to perform error prediction, and perform zero bias correction on the predicted value; S3-5, the corrected error prediction value and the inertial navigation recursive result are fused through an extended Kalman filter, wherein the observation noise covariance of the filter is dynamically adjusted according to the historical prediction accuracy of the current phase interval; S3-6 After the occlusion ends, for subnetworks whose prediction accuracy is lower than the preset accuracy threshold, the learning rate is increased to a predetermined multiple of the normal value for correction.

[0132] It should be noted that the rotor period refers to the time required for the rotor to rotate once, which is equal to the reciprocal of the rotor speed.

[0133] A phase interval refers to a series of sub-intervals obtained by dividing a complete rotor cycle into equal intervals according to the instantaneous phase. Each interval corresponds to a specific angular range during the rotor blade rotation process.

[0134] A REF neural network subnetwork is an independent instance of a radial basis function neural network, specifically designed to learn error characteristics within a specific phase interval. Each subnetwork has the same three-layer structure.

[0135] Radial basis function neural networks (RBNs) are three-layer feedforward neural networks that use Gaussian functions as activation functions. They consist of an input layer, a hidden layer, and an output layer and are used for approximating nonlinear functions.

[0136] The modulation depth threshold is a critical value used to distinguish between normal and occlusion states. When the modulation depth exceeds this threshold, it is determined to be occlusion.

[0137] Training refers to the process of using the position residuals of GNSS signals during normal periods as a supervisory signal and updating the weights of a neural network through a backpropagation algorithm.

[0138] Weights are adjustable parameters between connected nodes in a neural network, which determine the mapping relationship between input and output.

[0139] The obstruction state refers to the state in which the signal quality deteriorates when the rotor blades rotate to the direction of the line connecting the GNSS antenna and the satellite. In this state, the carrier phase observation value is unreliable.

[0140] The upper limit of continuous cycle count refers to the minimum number of consecutive cycles required to determine that a GNSS signal has entered a state of obstruction, in order to avoid misjudgment caused by transient noise.

[0141] Error prediction refers to the inertial navigation error estimate output by the REF neural network subnetwork based on the current input features, which is used to compensate for the divergence of inertial navigation.

[0142] Zero bias correction refers to subtracting the statistical mean of historical errors within the phase interval from the predicted error value of the subnetwork output in order to eliminate the systematic bias of the subnetwork.

[0143] Inertial navigation recursive results refer to the position, velocity, and attitude information obtained by integrating and recursively calculating the angular velocity and acceleration measured by gyroscopes and accelerometers.

[0144] The observation noise covariance is a matrix parameter in the extended Kalman filter that describes the uncertainty of the observations. The larger the covariance, the less reliable the observations are.

[0145] Historical prediction accuracy refers to the accuracy of error prediction by the REF neural network subnetwork over a period of time, which is determined by the statistical characteristics of the predicted value and the measured residual.

[0146] The learning rate is a parameter that specifies the step size for updating weights during the training of a neural network. The larger the learning rate, the faster the weights change.

[0147] The predetermined multiplier refers to the multiplier by which the learning rate of the low-precision subnetwork is increased after the occlusion ends, which is used to quickly correct the prediction bias of the subnetwork.

[0148] The normal period for GNSS signals refers to the time period when the modulation depth does not exceed the modulation depth threshold and the rotor does not obstruct the GNSS antenna.

[0149] Step S3 of this application overcomes the limitation of existing technologies where using a single neural network to learn the error characteristics of the entire cycle leads to mutual interference between different phase errors by dividing the rotor cycle into multiple intervals according to the instantaneous phase and establishing an independent REF neural network sub-network for each interval. Specifically, each rotor cycle is divided into K equally spaced phase intervals according to the instantaneous phase output of the phase-locked loop, where K is an integer multiple of the number of rotor blades, and an independent REF neural network sub-network is established for each interval. Since the degree of rotor obstruction of the GNSS antenna varies in different phase intervals, the error characteristics of inertial navigation also differ. After dividing the entire cycle, each sub-network only needs to learn the error pattern within a specific phase interval, avoiding mutual interference between different phases. When the GNSS signal is normal, only the sub-network corresponding to the current instantaneous phase is activated for training, while the weights of other sub-networks are frozen, allowing each sub-network to focus on learning the error characteristics of its own phase. In the obstruction state, the corresponding sub-network is activated to perform error prediction and zero-bias correction, and the predicted value is fused with the inertial navigation recursive result through an extended Kalman filter. The observation noise covariance of the filter is dynamically adjusted based on the historical prediction accuracy of the phase interval, with sub-networks having higher prediction accuracy receiving higher weights in the fusion process. After the occlusion ends, the learning rate of sub-networks with lower prediction accuracy is rapidly increased for correction. The entire process forms an adaptive neural network compensation chain that is phase-based, learns independently, and fuses as needed, enabling the split-type flying car to suppress inertial navigation divergence by utilizing historical learning experience when the rotor is occluded.

[0150] Step S4 further includes the following sub-steps: S4-1 maps the ratio of the instantaneous phase output by the phase-locked loop to the rotor speed to the blade angular position from 0 to 360 degrees, and calculates the uncertainty of the angular position based on the root mean square of the phase error of the phase-locked loop in the past rotor cycle. S4-2, a critical angle window is preset according to the number of rotor blades. The width of the critical angle window is 180 degrees divided by the number of blades. The center of the critical angle window is located at the position where the direction of the line connecting the rotor blade and the GNSS antenna coincides. S4-3, the difference between the angular position and its uncertainty is used as the lower confidence limit. When the lower confidence limit falls into the danger angle window multiple times in a row, it is determined that the rotor blade is about to sweep across the GNSS antenna's field of view. S4-4 calculates the basic risk coefficient based on the current altitude and descent speed of the flight module, and multiplies this coefficient by the root mean square of the phase error of the phase-locked loop as the corrected risk coefficient, increasing the safety domain radius to 1 times the original radius plus the corrected risk coefficient.

[0151] It should be noted that the root mean square of the phase error refers to the root mean square value of the phase tracking error of the phase-locked loop over the past rotor cycle, and is used to quantify the uncertainty of the phase measurement.

[0152] Uncertainty refers to the radius of the confidence interval for angular position calculated from the root mean square of the phase error, representing the reliable range of the blade angular position measurement.

[0153] The number of blades refers to the number of blades on a single rotor of a flight module, commonly two, three, or four blades.

[0154] The window width refers to the angular range of the danger angle window, which is equal to 180 degrees divided by the number of blades.

[0155] The center of the window refers to the middle position of the danger angle window, located at the angle where the direction of the line connecting the propeller blade and the GNSS antenna coincides.

[0156] The lower confidence limit refers to the difference between the angular position and its uncertainty as the criterion, representing the minimum possible value of the blade angular position after considering measurement errors.

[0157] "Successive times" refers to the number of times the lower confidence limit falls within the danger angle window consecutively, used to avoid misjudgments caused by single noise.

[0158] Sweeping across the antenna field of view refers to the state in which the rotor blades rotate to the direction of the line connecting the GNSS antenna and the satellite, which will block the signal.

[0159] The basic risk coefficient is a risk assessment value calculated based on the current altitude and descent speed of the flight module, reflecting the degree of danger of the current descent state.

[0160] The corrected risk coefficient is a correction value obtained by multiplying the basic risk coefficient by the root mean square of the phase error of the phase-locked loop, and is used to adjust the increase factor of the safety domain radius.

[0161] The original radius refers to the default safety domain radius under risk-free conditions.

[0162] Step S4 of this application overcomes the limitations of existing technologies that ignore phase measurement errors and use fixed angle thresholds to determine occlusions, which lead to misjudgments or missed judgments, by introducing an uncertainty calculation and confidence lower limit determination mechanism based on the root mean square of phase error. Specifically, the ratio of the instantaneous phase output by the phase-locked loop to the rotor speed is mapped to the blade angular position, and the uncertainty of the angular position is calculated based on the root mean square of the phase error over the past rotor cycle. This uncertainty reflects the reliability of the phase measurement; the larger the phase error, the larger the uncertainty and the wider the confidence interval. A danger angle window is preset based on the number of rotor blades, with a window width of 180 degrees divided by the number of blades, and the center of the window is located at the position where the blade and the GNSS antenna line coincide. The uncertainty is subtracted from the angular position to obtain the confidence lower limit. When the confidence lower limit falls into the danger angle window multiple times consecutively, it is determined that the rotor blade is about to sweep across the GNSS antenna's field of view. This determination mechanism takes into account measurement errors: when the phase measurement uncertainty is large, the confidence lower limit is lower, making it easier to trigger the determination, reflecting adaptability to measurement errors. Simultaneously, a basic risk coefficient is calculated based on the flight module's altitude and descent speed. This coefficient is then multiplied by the root mean square of the phase error to obtain a corrected risk coefficient, increasing the safety domain radius to one times the original radius plus the corrected risk coefficient. The larger the phase error, the greater the increase in the safety domain radius. This entire process forms a complete chain from uncertainty quantification to confidence lower limit determination to adaptive safety domain adjustment, enabling the split-type flying car to automatically increase its safety margin when measurements are unreliable.

[0163] Step S5 further includes the following sub-steps: S5-1 monitors the instantaneous phase of the phase-locked loop output in real time and constructs a phase sliding window with a length of N rotor cycles; S5-2, calculate the standard deviation of the phase within the window. When the standard deviation exceeds the preset standard deviation threshold, it is determined that the standard deviation condition is met. S5-3, Perform Fourier transform on the phase sequence within the window, extract random components in the spectrum that are unrelated to the rotor speed fundamental frequency and its harmonics, calculate the percentage of random component energy in the total energy, and determine that the random component condition is met when the percentage exceeds the preset percentage threshold. S5-4, when the standard deviation condition and the random component condition are satisfied at the same time, it is determined that the antenna phase center change is caused by the rotor airflow, and the instantaneous phase at the moment of change is recorded as the interference phase. S5-5, generate a pause docking command, stop the descent motion of the flight module, re-execute step S1, and predict the arrival time of the next airflow interference based on the occurrence pattern of the interference phase during re-execution, and use the recorded interference phase as the initial phase estimate of the phase-locked loop.

[0164] It should be noted that a phase sliding window refers to a fixed-length instantaneous phase sequence window that slides and updates in chronological order, used to analyze the statistical characteristics and spectral features of phase signals.

[0165] Standard deviation refers to the standard deviation of an instantaneous phase sequence within a phase sliding window, and is used to quantify the degree of phase fluctuation.

[0166] The preset standard deviation threshold is a critical value for judging whether the phase is abnormal. Exceeding this threshold indicates that the phase fluctuation is too large.

[0167] The rotor speed fundamental frequency refers to the basic frequency of rotor rotation, which is equal to the rotor speed multiplied by the number of blades.

[0168] Harmonics refer to components that are integer multiples of the fundamental frequency, generated by the periodic motion of the rotor.

[0169] Random components refer to frequency components in the spectrum that are independent of the rotor speed fundamental frequency and its harmonics, and are caused by non-periodic airflow disturbances.

[0170] The preset percentage threshold refers to the critical value of the random component energy percentage used to determine whether there is significant airflow disturbance.

[0171] The standard deviation condition refers to the state where the standard deviation of the phase sliding window exceeds the preset standard deviation threshold, indicating that the phase fluctuation is too large.

[0172] The random component condition refers to the state in which the percentage of random component energy in the total energy exceeds a preset percentage threshold, indicating the presence of non-periodic airflow disturbance.

[0173] Antenna phase center abrupt change refers to the phenomenon that the phase center of a GNSS antenna shifts instantaneously due to rotor airflow disturbance, manifesting as an abnormal jump in the instantaneous phase.

[0174] Interference phase refers to the instantaneous phase value recorded at the moment of abrupt change, which is used for subsequent airflow interference prediction and phase-locked loop acceleration locking.

[0175] The pause docking command is a control command that stops the descent of the flight module and pauses the docking process.

[0176] Descent motion refers to the process of the flight module descending vertically towards the ground module.

[0177] The occurrence pattern of interference phase refers to the time interval or periodicity of airflow interference, which is used to predict the arrival time of the next interference.

[0178] The initial phase estimate refers to the initial phase value used when the phase-locked loop restarts. Using historical interference phases as prior information can accelerate locking.

[0179] Step S5 of this application overcomes the limitations of existing technologies, such as the inability to identify aperiodic airflow interference and slow relocking speed after interference, by introducing a dual-condition determination mechanism based on the standard deviation of the sliding window and the energy of random components, as well as an interference phase recording and prediction feedback mechanism. Specifically, the instantaneous phase output of the phase-locked loop is monitored in real time, and a phase sliding window with a length of N rotor cycles is constructed. The standard deviation of the phase within the window is calculated, reflecting the severity of phase fluctuations. A Fourier transform is performed on the phase sequence within the window to extract random components in the spectrum that are unrelated to the rotor speed fundamental frequency and its harmonics, and the percentage of their energy in the total energy is calculated, reflecting the intensity of aperiodic airflow interference. When both the standard deviation condition and the random component condition are met simultaneously, it is determined to be a sudden change in the antenna phase center caused by rotor airflow. Only when both conditions are met simultaneously can it be confirmed as airflow interference, avoiding misjudging rotor speed fluctuations as airflow interference. The instantaneous phase at the moment of the sudden change is recorded as the interference phase, a pause docking command is generated, the descent motion is stopped, and step S1 is re-executed. Upon re-execution, the arrival time of the next airflow interference is predicted based on the occurrence pattern of the interference phase, allowing the system to prepare in advance. Simultaneously, the recorded interference phase is used as the initial phase estimate for the phase-locked loop (PLL), enabling the PLL to start tracking from near that phase instead of searching from zero, significantly shortening the relocking time. This entire process forms a closed-loop adaptive chain from phase monitoring, dual-condition determination, interference phase recording, docking pause, interference pattern prediction to accelerated relocking, allowing the split-type flying car to accurately identify airflow interference and recover quickly.

[0180] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A precise docking guidance method for a split-type flying car, characterized in that, Includes the following steps: Step S1: Acquire rotor speed signal, extract modulation component with the same frequency as rotor speed from GNSS carrier-to-noise ratio sequence, track its instantaneous phase through phase-locked loop, determine the obstruction phase according to modulation depth, and dynamically adjust the quality weight of carrier phase observation value. Step S2: Based on the occlusion phase, predict the signal interruption interval, lock the integer ambiguity before the interruption arrives, keep the ambiguity unchanged within the interruption interval, and only update the position parameters; Step S3: When the modulation depth exceeds the threshold for multiple consecutive cycles, the inertial navigation and visual pose are fused through an extended Kalman filter, and the REF neural network is trained with instantaneous phase to compensate for periodic errors. Step S4: At the end of the docking phase, the ratio of the instantaneous phase to the rotor speed is mapped to the blade angle position. When the angle position falls into the preset danger angle window multiple times in a row, the safety zone radius is increased. Step S5: When the sliding window standard deviation of the instantaneous phase and the random component energy both exceed the threshold, it is determined that the antenna change is caused by airflow, the docking is suspended and step S1 is re-executed.

2. The precise docking guidance method for a split-type flying car according to claim 1, characterized in that: Step S1 further includes the following sub-steps: S1-1, A vibration sensor is installed at the root of the rotor blade. The collected flapping vibration signal is subjected to Fourier transform, the fundamental frequency component is extracted as the rotor speed signal, and the rotor imbalance is judged from the sideband amplitude of the fundamental frequency component. S1-2, the rotor imbalance is determined by the amplitude ratio of the sideband component to the fundamental frequency component. When the imbalance exceeds the preset imbalance threshold, the loop bandwidth of the phase-locked loop is temporarily increased to a preset multiple of the fundamental frequency to suppress sideband interference. S1-3, adaptive filtering of GNSS carrier-to-noise ratio sequence, setting the center frequency of the filter to the rotor speed base frequency, and introducing a frequency traction auxiliary loop. When the instantaneous phase error of the phase-locked loop exceeds the preset angle threshold, the center frequency is pulled back to the preset frequency range of the rotor speed base frequency through the frequency traction auxiliary loop. S1-4: Using the phase difference output by the phase-locked loop phase detector and the flapping phase of the rotor blades, calculate the dynamic deformation of the rotor blades in the plane of rotation. When the dynamic deformation exceeds the preset proportional threshold of the blade length, it is determined that the rotor has generated elastic deformation. S1-5, calculate the correction amount based on the magnitude of the rotor dynamic deformation and the blade length, and correct the spatial angle between the rotor blade and the GNSS antenna line. The correction amount is the ratio of the deformation to the blade length multiplied by a preset angle coefficient. S1-6, calculate the modulation depth based on the filtered GNSS carrier-to-noise ratio sequence, determine the obstruction phase based on the magnitude of the modulation depth, and adjust the quality weight of the carrier phase observation value in stages. The modulation depth is the ratio of the peak value of the carrier-to-noise ratio to the mean value of the carrier-to-noise ratio. S1-7, within the unobstructed phase interval, the instantaneous phase of the phase-locked loop is smoothed using the carrier phase observation value. The smoothed phase is used to correct the phase prediction value in the obstructed interval of the next cycle, forming a closed-loop correction of the phase prediction.

3. The precise docking guidance method for a split-type flying car according to claim 2, characterized in that, In sub-steps S1-6, determining the blocking phase based on the modulation depth includes: determining the number of sampling points in a single rotor cycle based on the rotor speed; using the proportion of sampling points with a modulation depth exceeding 3dB in a single rotor cycle exceeding 70% as the starting condition for the blocking phase; and using the proportion of this proportion being less than 30% in two consecutive rotor cycles as the ending condition for the blocking phase.

4. The precise docking guidance method for a split-type flying car according to claim 2, characterized in that, In sub-steps S1-6, the graded adjustment includes: calculating the real-time angular distance between the rotor blade and the GNSS antenna based on the instantaneous phase output by the phase-locked loop; initiating pre-weight reduction when the angular distance is less than 30 degrees, reducing the quality weight of the carrier phase observation value to 50% of the target weight; and reducing it to the target weight when the angular distance is less than 10 degrees, thus forming a continuous two-level weight reduction mechanism related to the angular distance.

5. The precise docking guidance method for a split-type flying car according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1, Based on the instantaneous phase output by the phase-locked loop and the rotor speed, combined with the number of rotor blades, calculate the predicted blocking phase at each moment in the next two rotor cycles, and calculate the prediction error covariance of the blocking phase. The prediction error covariance is the square of the instantaneous standard deviation of the rotor speed multiplied by a preset proportional coefficient. S2-2: Based on the deviation between the predicted blocking phase and the measured blocking phase of the previous cycle, determine the rotor rotation state. If it is determined to be in a steady state, execute S2-3 to S2-6. If it is determined to be in a transient state, skip S2-3 to S2-6, switch to pure inertial navigation mode, and continue to run pure inertial navigation mode until the instantaneous standard deviation of rotor speed recovers to below the preset standard deviation threshold, and start executing again from S2-1. S2-3, before the interruption start time arrives, normalize the prediction error covariance to the [0,1] interval, and adjust the elevation angle threshold in the ambiguity fixing strategy according to the magnitude of the prediction error covariance, so that the threshold changes linearly between 15 degrees and 45 degrees. The specific formula is T=15+30C, where C is the normalized covariance value. S2-4: Store the fixed integer ambiguity value into a non-volatile register and freeze the register during the interrupt interval; S2-5, within the interruption interval, using the carrier phase position solution at the start of the interruption as the initial value, the position is updated by using the velocity increment recursively derived by inertial navigation, keeping the integer ambiguity unchanged; S2-6 After the interruption interval ends, the difference between the initial carrier phase value recorded at the start time of the interruption interval and the final carrier phase value recorded at the end time of the interruption interval is used to calculate the integer ambiguity change in the interruption interval. When the absolute value of the back-calculated change exceeds the preset change threshold, it is determined that a cycle slip has occurred in the interruption interval, triggering ambiguity re-fixing.

6. The precise docking guidance method for a split-type flying car according to claim 5, characterized in that, In sub-step S2-2, the step of determining the rotor rotation state includes: Calculate the absolute value of the deviation between the predicted occlusion phase and the measured occlusion phase of the previous cycle, and obtain the standard deviation of the prediction error; When the absolute value of the deviation is less than 0.5 times the standard deviation of the prediction error for three consecutive times, it is determined that the rotor has entered a steady state. At this time, the predicted occlusion phase is reliable, and the interruption prediction mode is started. Otherwise, it is judged as transient, in which case the prediction is unreliable, and interruption of the prediction is prohibited.

7. The precise docking guidance method for a split-type flying car according to claim 5, characterized in that, In sub-steps S2-6, the step of triggering ambiguity refixing includes: The last integer ambiguity before the interruption interval is used as prior information, and it is converted into a pseudo-observation value and added to the double-difference observation equation to constrain the calculation of the floating-point ambiguity solution; By using the position change recursively calculated within the interruption interval as a geometric constraint, the ambiguity search space is reduced; The LAMBDA algorithm is used to fix the integer ambiguity, and only the subset of satellites whose absolute value of the back-calculated change exceeds the preset change threshold is re-fixed, while the integer ambiguity of the remaining satellites remains unchanged from the fixed value before the interruption.

8. The precise docking guidance method for a split-type flying car according to claim 1, characterized in that: Step S3 further includes the following sub-steps: S3-1, Based on the instantaneous phase output by the phase-locked loop, each rotor cycle is divided into K equally spaced phase intervals, and an independent REF neural network sub-network is established for each phase interval, where K is an integer multiple of the number of rotor blades; S3-2, when the modulation depth does not exceed the modulation depth threshold, activate the corresponding sub-network according to the current instantaneous phase for training and update the weights of the sub-network; S3-3, Set the upper limit of continuous cycle count. When the number of cycles in which the modulation depth continuously exceeds the modulation depth threshold reaches or exceeds the upper limit, it is determined that the GNSS signal is in a blocked state. S3-4, In the occlusion state, activate the corresponding sub-network according to the current instantaneous phase to perform error prediction, and perform zero bias correction on the predicted value; S3-5, the corrected error prediction value and the inertial navigation recursive result are fused through an extended Kalman filter, wherein the observation noise covariance of the filter is dynamically adjusted according to the historical prediction accuracy of the current phase interval; S3-6 After the occlusion ends, for subnetworks whose prediction accuracy is lower than the preset accuracy threshold, the learning rate is increased to a predetermined multiple of the normal value for correction.

9. The precise docking guidance method for a split-type flying car according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1 maps the ratio of the instantaneous phase output by the phase-locked loop to the rotor speed to the blade angular position from 0 to 360 degrees, and calculates the uncertainty of the angular position based on the root mean square of the phase error of the phase-locked loop in the past rotor cycle. S4-2, a critical angle window is preset according to the number of rotor blades. The width of the critical angle window is 180 degrees divided by the number of blades, and the center of the critical angle window is located at the position where the direction of the line connecting the blade and the GNSS antenna coincides. S4-3, the difference between the angular position and its uncertainty is used as the lower confidence limit. When the lower confidence limit falls into the danger angle window multiple times in a row, it is determined that the rotor blade is about to sweep across the GNSS antenna's field of view. S4-4 calculates the basic risk coefficient based on the current altitude and descent speed of the flight module, and multiplies this coefficient by the root mean square of the phase error of the phase-locked loop as the corrected risk coefficient, increasing the safety domain radius to 1 times the original radius plus the corrected risk coefficient.

10. The precise docking guidance method for a split-type flying car according to claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1 monitors the instantaneous phase of the phase-locked loop output in real time and constructs a phase sliding window with a length of N rotor cycles; S5-2, calculate the standard deviation of the phase within the window. When the standard deviation exceeds the preset standard deviation threshold, it is determined that the standard deviation condition is met. S5-3, Perform Fourier transform on the phase sequence within the window, extract random components in the spectrum that are unrelated to the rotor speed fundamental frequency and its harmonics, calculate the percentage of random component energy in the total energy, and determine that the random component condition is met when the percentage exceeds the preset percentage threshold. S5-4, when the standard deviation condition and the random component condition are satisfied at the same time, it is determined that the antenna phase center change is caused by the rotor airflow, and the instantaneous phase at the moment of change is recorded as the interference phase. S5-5, generate a pause docking command, stop the descent motion of the flight module, re-execute step S1, and predict the arrival time of the next airflow interference based on the occurrence pattern of the interference phase during re-execution, and use the recorded interference phase as the initial phase estimate of the phase-locked loop.

Images