Crossed dual-rotor unmanned aerial vehicle adaptive stability control system based on gust perception and trend prediction

Abstract

The application belongs to the technical field of unmanned aerial vehicle flight control and multi-source data processing, and particularly relates to a cross dual-rotor unmanned aerial vehicle adaptive stability control system based on gust perception and trend prediction, which comprises the following steps: multi-source flight state time series data characteristics are obtained; rotor load fluctuation characteristics and corresponding wind speed change rate and wind direction change rate of micro-meteorological time series data are subjected to feature splicing and normalization processing to generate a multi-dimensional gust feature vector; the multi-dimensional gust feature vector is input into a time series prediction network to output a disturbance risk level; based on the disturbance risk level, a gain parameter matrix of a flight controller is dynamically reconstructed, and a feedforward compensation moment is calculated based on a wind field disturbance state vector; a front and rear rotor thrust distribution control instruction is generated and output to a motor drive unit. The application can solve the problems of lack of multi-source time series data frequency domain feature extraction, wind field trend feedforward prediction and thrust distribution imbalance in edge abnormal scenarios in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of UAV flight control and multi-source data processing technology, specifically relating to an adaptive stability control system for a cross-rotor UAV based on gust perception and trend prediction. Background Technology

[0002] With the rapid development of UAV technology, it has been widely used in power line inspection, wind power maintenance, logistics transportation, forest fire prevention, emergency rescue, and low-altitude inspection. However, in mountainous canyons, urban building complexes, wind farms, and complex terrain environments, airflow is easily affected by terrain, buildings, and obstacles, forming unsteady gusts and local turbulent structures, leading to attitude fluctuations, trajectory deviations, and decreased flight stability during UAV flight. Cross-rotor UAVs have significant advantages in complex environment missions due to their high lift efficiency, strong wind resistance, good hovering stability, and large payload capacity. However, in actual flight, gust disturbances can cause transient changes in rotor lift, resulting in additional attitude deviations and vibration responses, thus affecting flight control accuracy and mission execution. Most existing UAV control systems use fixed-parameter PID control, gain scheduling control, or empirical compensation methods, whose control parameters rely on pre-set parameters and are difficult to adapt to rapidly changing wind field environments in real time.

[0003] Problems with existing technology: However, when encountering high-frequency gust disturbances, the existing system lacks a frequency domain feature extraction mechanism for multi-source heterogeneous time-series data and a feedforward prediction mechanism for wind field evolution trends, resulting in the control gain being unable to dynamically match transient aerodynamic load mutations. Furthermore, in edge anomaly scenarios such as motor output saturation or sensor data loss, the fixed allocation matrix is ​​prone to causing rotor anti-torque imbalance and state estimation divergence, leading to flight instability. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive stability control system for a cross-rotor UAV based on gust perception and trend prediction, which can solve the problems of lack of multi-source time-series data frequency domain feature extraction, wind field trend feedforward prediction, and thrust distribution imbalance in edge anomaly scenarios in the prior art.

[0005] The specific technical solution adopted by this invention is as follows: An adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction includes the following steps: In response to acquiring multi-source flight status time-series data and micro-meteorological time-series data of the cross-rotor UAV, a short-time Fourier transform is performed on the body vibration signal and rotor speed signal in the multi-source flight status time-series data to extract rotor load fluctuation characteristics. The rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data for feature splicing and normalization to generate a multi-dimensional gust feature vector. The multidimensional gust feature vector is input into a pre-trained time-series prediction network, which outputs the wind field disturbance state vector and the corresponding disturbance risk level within a future preset time window. Based on the disturbance risk level, the gain parameter matrix of the flight controller is dynamically reconstructed, and the feedforward compensation torque is calculated based on the wind field disturbance state vector. By combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV and output to the motor drive unit.

[0006] According to another aspect of the present invention, a short-time Fourier transform is performed on the airframe vibration signal and rotor speed signal in the multi-source flight state time-series data to extract rotor load fluctuation characteristics, including: The airframe vibration signal and the rotor speed signal are windowed and truncated using a Hanning window of a preset length, and sliding sampling is performed according to a preset overlap rate to obtain multiple local time-frequency segments; Perform a fast Fourier transform on each of the local time-frequency segments to extract the energy spectral density of the corresponding frequency band; Based on the peak amplitude of the energy spectral density in the rotor fundamental frequency and its harmonic frequency bands, the fluctuation variance of the rotor aerodynamic load is calculated, and the fluctuation variance is used as the rotor load fluctuation characteristic.

[0007] According to another aspect of the present invention, the rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data for feature concatenation and normalization to generate a multi-dimensional gust feature vector, including: Calculate the difference of the micro-meteorological time series data within a continuous sampling period to obtain the wind speed change rate and the wind direction change rate; The rotor load fluctuation characteristics, the wind speed change rate, the wind direction change rate, and the turbulence intensity index are concatenated into a one-dimensional initial feature vector; The mean and standard deviation of each dimension in the one-dimensional initial feature vector are calculated based on the historical wind field dataset. The one-dimensional initial feature vector is then normalized using the Z-score normalization algorithm to output the multi-dimensional gust feature vector.

[0008] According to another aspect of the present invention, the time-series prediction network is a Transformer model containing a multi-head self-attention mechanism. The multi-dimensional gust feature vector is input into the pre-trained time-series prediction network, and the network outputs a wind field disturbance state vector within a preset future time window and the corresponding disturbance risk level, including: The multidimensional gust feature vector is mapped into a query vector, a key vector, and a value vector; The attention weight matrix is ​​obtained by calculating the dot product of the query vector and the key vector, dividing it by the scaling factor, and then processing it through the Softmax function. The spatiotemporal evolution features of the wind field are extracted by weighting and summing the value vector using the attention weight matrix. The spatiotemporal evolution features of the wind field are input into a fully connected classification layer, the wind field disturbance state vector is output, and the disturbance risk level is matched and output based on the disturbance intensity threshold and duration threshold in the wind field disturbance state vector. The disturbance risk levels include five levels: no disturbance, slight, moderate, severe, and extremely severe. Based on the disturbance risk level, the gain parameter matrix of the flight controller is dynamically reconstructed, including: When the disturbance risk level is no disturbance or slight disturbance, the inference process of the time series prediction network is bypassed, and the stiffness gain, damping gain and integral gain of the flight controller are determined by calling the preset basic gain parameter table. When the disturbance risk level is medium, the damping gain in the basic gain parameter table is amplified and compensated based on the first weighting coefficient. When the disturbance risk level is severe or extremely severe, the gain adaptive adjustment is triggered. Based on the second weighting coefficient, the stiffness gain and damping gain in the basic gain parameter table are amplified and compensated, and an anti-interference weighting factor is introduced to attenuate the integral gain, thereby generating the reconstructed gain parameter matrix.

[0009] According to another aspect of the present invention, calculating the feedforward compensation torque based on the wind field disturbance state vector includes: A wind field disturbance model is established based on the wind field disturbance state vector to analyze the three-axis disturbance force and three-axis disturbance torque experienced by the cross-rotor UAV in the body coordinate system. Based on the three-axis disturbance force and the three-axis disturbance torque, the attitude compensation amount and trajectory compensation amount are calculated respectively. The trajectory compensation amount is converted into a desired attitude command through the outer loop position controller, and the attitude compensation amount is converted into a desired angular velocity through the inner loop attitude controller; Based on the desired attitude command and the desired angular velocity, inverse dynamics calculation is performed to generate the feedforward compensation torque.

[0010] According to another aspect of the present invention, by combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV, including: The feedback control torque is calculated based on the current attitude error of the UAV and the reconstructed gain parameter matrix. The feedback control torque is superimposed with the feedforward compensation torque to obtain the total demand control torque; Construct a thrust distribution matrix that includes the thrust coefficients and anti-torque coefficients of the front and rear rotors, and calculate the total demand control torque based on the pseudo-inverse matrix of the thrust distribution matrix to obtain the basic thrust command of the front and rear rotors. Based on the cross-coupled aerodynamic interference model of the front and rear rotors, the basic thrust command is corrected by anti-torque compensation to generate the thrust distribution control command.

[0011] According to another aspect of the present invention, it also includes exception handling logic for motor output saturation: If any rotor thrust value in the thrust distribution control command exceeds the maximum physical thrust threshold of the motor, the thrust redistribution mechanism is triggered. The rotor thrust value exceeding the maximum physical thrust threshold of the motor is truncated to the maximum physical thrust threshold of the motor, and the thrust loss is calculated. Based on the lever arm ratio of the cross-rotor UAV, the thrust deficit is proportionally allocated to the unsaturated rotor, and the anti-torque compensation coefficient of the other rotor is adjusted simultaneously to update the thrust distribution control command.

[0012] According to another aspect of the present invention, degradation is also included for sensor data packet loss or timeout: If continuous packet loss is detected in the micro-meteorological time-series data or the multi-source flight status time-series data, and the packet loss duration exceeds a preset timeout threshold, the inference of the time-series prediction network is suspended. The extended Kalman filter is invoked, and the flight state and wind field disturbance state at the current moment are recursively estimated based on the state vector and system state transition matrix of the last valid moment before packet loss. The gain parameter matrix of the flight controller is downgraded to a fixed parameter mode, and the feedforward compensation torque is calculated based on the recursively estimated wind field disturbance state.

[0013] According to another aspect of the present invention, degradation processing for edge computing resource constraints is also included: If the utilization rate of the flight control main processor executing the method exceeds a preset utilization rate threshold for multiple consecutive control cycles, model degradation is triggered. The time series prediction network is switched to a gated recurrent unit network with fewer parameters than the time series prediction network, or a three-dimensional lookup table of wind field characteristics and control gain generated offline based on historical wind field data is called to perform interpolation calculations to update the gain parameter matrix.

[0014] According to another aspect of the present invention, an adaptive stability control system for a cross-rotor unmanned aerial vehicle based on gust perception and trend prediction is also provided, comprising: The feature extraction and fusion module is configured to, in response to acquiring multi-source flight state time-series data and micro-meteorological time-series data of the cross-rotor UAV, perform short-time Fourier transform on the body vibration signal and rotor speed signal in the multi-source flight state time-series data to extract rotor load fluctuation features. The rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data for feature splicing and normalization to generate a multi-dimensional gust feature vector. The trend prediction module is configured to input the multi-dimensional gust feature vector into a pre-trained time-series prediction network and output the wind field disturbance state vector and the corresponding disturbance risk level within a future preset time window. The control reconfiguration and allocation module is configured to dynamically reconfigure the gain parameter matrix of the flight controller based on the disturbance risk level, and calculate the feedforward compensation torque based on the wind field disturbance state vector. By combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV and output to the motor drive unit.

[0015] According to another aspect of the present invention, an electronic device is also provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the adaptive stabilization control method for a cross-rotor unmanned aerial vehicle based on gust perception and trend prediction as described above.

[0016] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium storing computer instructions for causing a computer to execute the adaptive stabilization control method for a cross-rotor unmanned aerial vehicle based on gust perception and trend prediction as described above.

[0017] The technical effects achieved by this invention are as follows: This invention extracts rotor load fluctuation features by performing short-time Fourier transform on the body vibration signal and rotor speed signal, and generates a multi-dimensional gust feature vector by splicing and normalizing the features with micro-meteorological time series data. This expands the input data for wind field perception from a single physical scalar to a multi-dimensional tensor that includes frequency domain energy spectrum and meteorological change rate, thereby enabling the cross-rotor UAV to accurately capture high-frequency gust disturbance features and accurately calculate feedforward compensation torque.

[0018] This invention employs a time-series prediction network with a multi-head self-attention mechanism to process multi-dimensional gust feature vectors and dynamically reconstructs the gain parameter matrix of the flight controller based on the output disturbance risk level. This enables the stiffness gain and damping gain of the control system to adaptively match the spatiotemporal evolution characteristics of the wind field within a future preset time window, thereby suppressing attitude overshoot and integral saturation oscillation caused by fixed control parameters under strong gust conditions.

[0019] This invention, by truncating the maximum physical thrust threshold of the motor and proportionally redistributing the thrust deficit, and by calling an extended Kalman filter for state recursive estimation when sensor data is continuously lost, ensures that the control commands of the system still satisfy the rotor anti-torque balance constraint when the edge actuator is saturated or the communication link is abnormal. This avoids thrust imbalance and state estimation divergence in cross-rotor UAVs under extreme conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the flight status perception structure of the cross-rotor UAV of the present invention; Figure 2 This is a block diagram of the overall system structure of the present invention; Figure 3 This is a flowchart of an adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to the present invention. Figure 4 This is a flowchart of the gust sensing module of the present invention; Figure 5 This is a flowchart of the gust disturbance identification module of the present invention; Figure 6 This is a schematic diagram of the working principle of the environmental disturbance compensation module of the present invention; Figure 7 This is a flowchart of the gust trend prediction process of the present invention; Figure 8 This is the flowchart of the adaptive stability control of the present invention; Figure 9 This is a schematic diagram of the flight stability control principle of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0022] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] According to an embodiment of the present invention, a method embodiment of an adaptive stabilization control method for a cross-rotor unmanned aerial vehicle based on gust perception and trend prediction is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] In one optional embodiment, an adaptive stability control system for a cross-rotor UAV based on gust perception and trend prediction (hereinafter referred to as the system) is provided. This system is deployed on the airborne flight control computing platform of the cross-rotor UAV.

[0025] As a preferred approach, the flight control computing platform adopts a heterogeneous multi-core architecture, including a microcontroller as the main controller and an FPGA (Field Programmable Gate Array) chip as a coprocessor.

[0026] See attached document Figure 1 Furthermore, the sensor array of the perception and recognition layer adopts a distributed spatial topology layout. Front / rear rotor speed sensors, IMU inertial measurement units, and barometric altimeters are deployed on the top of the UAV and near the rotors to acquire high-frequency rigid body motion data. Tail attitude sensors and body vibration sensors are deployed at the tail and center of gravity to capture the aerodynamic coupling high-frequency vibration unique to cross-rotor systems. Downward-looking vision / LiDAR, millimeter-wave radar, and forward-looking vision cameras are deployed at the bottom and front of the fuselage to construct a local environmental perception field. GNSS antennas and miniature wind speed and direction sensors are deployed in the unobstructed area outside the fuselage. Data from multiple heterogeneous sensors are aggregated to the flight control computing platform via CAN-FD bus and SPI high-speed interface.

[0027] The specific sensor layout corresponds to: The front rotor speed sensor, rear rotor speed sensor, IMU inertial measurement unit, and barometric altimeter are arranged in the top / rotor-related area. Tail attitude sensor, body vibration sensor, and body temperature and humidity sensor are arranged in the fuselage / tail area; The bottom / front area is equipped with a downward vision / LiDAR, a millimeter-wave radar, and a forward vision camera; GNSS antennas and wind speed and direction sensors are deployed in external / general areas.

[0028] In terms of data flow, the high-frequency raw data acquired by the sensor is directly written into the on-chip BRAM (Block RAM) of the FPGA via the DMA (Direct Memory Access) controller, bypassing the CPU cache and thus ensuring microsecond-level data throughput latency. The FPGA is responsible for performing low-level signal filtering and Fast Fourier Transform (FFT), and transmits the processed feature vectors to the SRAM of the ARM host via the AXI bus for use by the upper-level timing prediction network and control law solving module.

[0029] See attached document Figure 2 As an optional embodiment, the overall system structure is divided into a perception and recognition layer and a control and execution layer. The perception and recognition layer includes: The flight status perception module includes, but is not limited to, an IMU inertial measurement unit, an accelerometer, a GNSS positioning unit, a barometric altimeter, and a rotor speed detection unit, and is used to output status data. The gust sensing module includes a wind speed detection unit, a wind direction detection unit, a wind field change rate calculation unit, and a gust feature extraction unit, which are used to output wind field data. The gust disturbance identification module includes a flight state change analysis unit, a vibration spectrum analysis unit, a rotor load analysis unit, and a disturbance level judgment unit, which are used to output disturbance information; The gust trend prediction module includes a historical wind field database, a feature fusion analysis unit, a wind field prediction model unit, and a disturbance development trend assessment unit, which are used to output trend predictions.

[0030] The control and execution layer includes: The environmental disturbance compensation module includes a wind field disturbance modeling unit, a disturbance compensation calculation unit, an attitude correction unit, and a trajectory correction unit, which are used to output the compensation amount. The adaptive stability control module includes a thrust distribution unit, an attitude control unit, a gain self-tuning unit, a rotor coordination control unit, and a stability optimization unit, which are used to output control commands. The flight execution module includes a motor drive unit, a rotor actuator, and a flight control output unit.

[0031] The system data flow is as follows: after the external environment and flight mission inputs are processed by the perception and recognition layer, wind field data, disturbance information and trend prediction are generated, which are then input to the control and execution layer for compensation calculation and command generation, and finally drive the cross-rotor UAV to perform flight missions.

[0032] As an optional embodiment, refer to the appendix Figure 3 An adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction includes the following steps: S1. In response to acquiring multi-source flight status time-series data and micro-meteorological time-series data of the cross-rotor UAV, perform short-time Fourier transform on the body vibration signal and rotor speed signal in the multi-source flight status time-series data to extract rotor load fluctuation characteristics. S2. The rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data to perform feature splicing and normalization to generate a multi-dimensional gust feature vector. S3. Input the multi-dimensional gust feature vector into the pre-trained temporal prediction network and output the wind field disturbance state vector and the corresponding disturbance risk level within the future preset time window. S4. Based on the disturbance risk level, dynamically reconstruct the gain parameter matrix of the flight controller and calculate the feedforward compensation torque based on the wind field disturbance state vector. S5. Combining the reconstructed gain parameter matrix and feedforward compensation torque, generate control commands for the thrust distribution of the front and rear rotors of the cross-rotor UAV and output them to the motor drive unit.

[0033] As one implementation of an optional embodiment, regarding the aforementioned data flow and method process, when the system performs adaptive stability control, in response to acquiring multi-source flight status time-series data and micro-meteorological time-series data of the cross-rotor UAV, it initiates the gust perception and feature extraction process. The multi-source flight status time-series data includes IMU six-axis data (three-axis acceleration and three-axis angular velocity) acquired at a sampling rate of 200Hz and front and rear rotor speed signals acquired at a sampling rate of 100Hz; the micro-meteorological time-series data includes three-dimensional wind speed and wind direction data acquired at a sampling rate of 50Hz.

[0034] As an optional embodiment, refer to the appendix Figure 4 For the aforementioned micro-meteorological time series data, when performing gust sensing, the system performs sliding window mid-range filtering on the collected raw wind speed and wind direction data to remove pulse anomalies. Then, it calculates the wind speed change rate and wind direction change rate, and extracts the wind speed fluctuation intensity, wind direction fluctuation intensity, and turbulence intensity calculated based on wind speed variance, and outputs standardized wind field characteristic quantities.

[0035] Furthermore, the specific workflow of the gust sensing module includes: Start; collect wind speed and direction data; calculate the rate of change of wind speed. ; Calculate the rate of change of wind direction d ; Filtering and outlier removal; Extraction of gust features (wind speed fluctuation intensity) Wind direction fluctuation intensity turbulence intensity Output wind field characteristics; End.

[0036] As an optional embodiment, refer to the appendix Figure 5 The specific algorithm flow for extracting rotor load fluctuation characteristics by performing short-time Fourier transform on the airframe vibration signal and rotor speed signal is as follows: The airframe vibration signal and rotor speed signal are truncated by using a Hanning window of preset length, and sliding sampling is performed according to a preset overlap rate to obtain multiple local time-frequency segments. The Hanning window can effectively suppress the phenomenon of spectrum leakage.

[0037] Perform a fast Fourier transform on each local time-frequency segment to extract the energy spectral density of the corresponding frequency band.

[0038] Based on the peak amplitude of the energy spectral density at the rotor fundamental frequency and its harmonic frequency bands, the fluctuation variance of the rotor aerodynamic load is calculated, and the fluctuation variance is used as the characteristic of rotor load fluctuation.

[0039] The formulas for calculating the short-time Fourier transform (STFT) and load fluctuation variance are as follows:

[0040]

[0041] in, This represents the time-domain signal of the input airframe vibration or rotor speed; Indicates the center is The Hanning window function; This indicates the sampling point length of the Hanning window. In actual engineering deployments, to balance frequency resolution and time resolution, for a sampling rate of 200Hz, its empirical value is usually limited to 128 or 256. Indicates frequency domain index, Used to locate the rotor's fundamental frequency (i.e., the frequency corresponding to the rotor's rotational speed) and its second and third harmonic frequency bands; This represents the energy spectral density of the corresponding frequency band; This indicates the total number of frames in the sliding window; This represents the mean of the fundamental frequency energy spectral density; This refers to the variance of the output rotor aerodynamic load fluctuation, which physically reflects the degree of dispersion of the periodic excitation force generated by the airflow pulsation on the rotor blades at this frequency band.

[0042] Furthermore, the workflow of the gust disturbance identification module specifically includes: Start; Input flight status data and wind field characteristics; Flight status change analysis (attitude angle, angular velocity, acceleration); Vibration spectrum analysis (FFT / STFT); Rotor load fluctuation analysis; Feature fusion and normalization processing; Disturbance level judgment and direction identification; Output disturbance level and direction; End.

[0043] The disturbance levels are divided into: Lv.0 No disturbance, Lv.1 Slight, Lv.2 Moderate, Lv.3 Severe, and Lv.4 Extremely severe.

[0044] Furthermore, the system performs feature concatenation and normalization on the rotor load fluctuation characteristics and the corresponding wind speed and direction change rates from the micro-meteorological time-series data to generate a multi-dimensional gust feature vector, including: The rotor load fluctuation characteristics, wind speed change rate, wind direction change rate, and turbulence intensity index are concatenated into a one-dimensional initial feature vector (e.g., a 1×12 floating-point array). Subsequently, the mean and standard deviation of each dimension in the one-dimensional initial feature vector are calculated based on historical wind field datasets. The vector is then normalized using the Z-score normalization algorithm to eliminate the risk of gradient explosion caused by different physical dimensions, and a multi-dimensional gust feature vector is output.

[0045] As an optional embodiment, refer to the appendix Figure 6 It should be noted that since the six degrees of freedom of the cross-rotor UAV still has pain points under complex wind fields, the following environmental disturbance compensation is needed to solve this problem: Based on wind field characteristics and disturbance levels, the system establishes a wind field disturbance model that includes gusts and turbulence components, and analyzes the three-axis disturbance forces experienced by the UAV in the body coordinate system. With three-axis disturbance torque .

[0046] Based on the above model, the system calculates the attitude compensation amount respectively. With trajectory compensation amount In the control command generation stage, the trajectory compensation quantity is converted into the desired attitude command through the outer loop position controller, and the attitude compensation quantity is converted into the desired angular velocity through the inner loop attitude controller. Finally, it is uniformly converted into feedforward compensation torque through inverse dynamics calculation, realizing the transformation from macroscopic trajectory correction to microscopic torque compensation.

[0047] See attached document Figure 6 As shown in the diagram, the input information for the environmental disturbance compensation module includes: wind field characteristics, disturbance level and direction, flight status, and historical error data. The core processing includes: wind field disturbance modeling (based on gust + turbulence model), and disturbance effect model analysis (force). With torque ), Disturbance compensation calculation (attitude compensation amount) , , With trajectory compensation amount The final output information is: attitude correction command and trajectory correction command.

[0048] As an optional embodiment, refer to the appendix Figure 7 At the algorithm and model implementation level, the preferred temporal prediction network is a lightweight Transformer model incorporating a multi-head self-attention mechanism. The multi-dimensional gust feature vector is input into the pre-trained temporal prediction network, which outputs a wind field disturbance state vector and its corresponding disturbance risk level within a preset future time window (e.g., 0.5 to 2 seconds). Specifically, the multi-dimensional gust feature vector is transformed into a query vector, a key vector, and a value vector through a linear mapping layer. The attention weight matrix is ​​obtained by calculating the dot product of the query vector and the key vector, dividing by a scaling factor, and then processing with a Softmax function. The attention weight matrix is ​​used to perform a weighted summation of the value vectors to extract the spatiotemporal evolution features of the wind field. These features are then input into a fully connected classification layer, which outputs a wind field disturbance state vector.

[0049] The specific formula for calculating the multi-head self-attention mechanism is as follows:

[0050] in, These represent the query matrix, key matrix, and value matrix obtained by linear mapping of multidimensional gust feature vectors, respectively. The dimension of the key vector is used as the denominator of the scaling factor to prevent the dot product result from being too large and causing the Softmax function to enter the gradient saturation region. When the model pruning is deployed on an airborne embedded platform, its empirical value is usually 64 or 32. This represents the normalized exponential function used to output the attention weight distribution. When performing tasks involving navigating between urban buildings, to prevent crashes caused by sudden wind shear, it is possible to... A mask matrix is ​​introduced after calculation to shield invalid features at future time moments, thereby improving the system's sensitivity to moderate risks.

[0051] The gust trend forecasting process specifically includes: Start; Acquire historical wind field and disturbance data; Data preprocessing and feature extraction; Establish a wind field prediction model (LSTM / GRU / Transformer); Input real-time wind field features; Predict future wind field change trends; Assess disturbance development trends and risks; Output prediction results and risk levels; End.

[0052] The risk levels are divided into: R0 low risk, R1 moderate risk, R2 medium risk, R3 relatively high risk, and R4 high risk.

[0053] As an optional embodiment, refer to the appendix Figure 5 and Figure 7 In terms of disturbance level and risk level settings, the system adopts a five-level classification standard to match different control strategies.

[0054] The disturbance levels are categorized as follows: Lv.0 (no disturbance, stable airflow), Lv.1 (slight, gusts <3 m / s), Lv.2 (moderate, 3 m / s ≤ gusts < 6 m / s and turbulence intensity < 0.15), Lv.3 (severe, 6 m / s ≤ gusts < 10 m / s or significant airflow shearing), and Lv.4 (extremely severe, gusts ≥ 10 m / s or encountering extreme downbursts). The risk levels of the trend forecast output are categorized from R0 (low risk) to R4 (high risk).

[0055] It should be noted that the aforementioned "first risk level to fourth risk level" correspond to Lv.1 / R1 to Lv.4 / R4 states here, respectively. When the system determines that it is in the Lv.0 / R0 state, the flight control computing platform bypasses the inference process of the timing prediction network and directly calls the basic gain parameter table to save computing power at the edge.

[0056] Furthermore, based on the disturbance risk level, the gain parameter matrix of the flight controller is dynamically reconstructed. When the disturbance risk level is the first or second risk level, the preset basic gain parameter table is called to determine the stiffness gain, damping gain, and integral gain of the flight controller. When the disturbance risk level is the third or fourth risk level, adaptive gain adjustment is triggered. Based on the weight coefficients corresponding to the disturbance risk level, the stiffness gain and damping gain in the basic gain parameter table are amplified and compensated, and an anti-interference weight factor is introduced to attenuate the integral gain, generating the reconstructed gain parameter matrix.

[0057] Furthermore, by combining the reconstructed gain parameter matrix and feedforward compensation torque, control commands for the thrust distribution between the front and rear rotors of the cross-rotor UAV are generated, including: The feedback control torque is calculated based on the current attitude error of the UAV and the reconstructed gain parameter matrix. The total demand control torque is obtained by superimposing the feedback control torque and the feedforward compensation torque. Construct a thrust distribution matrix that includes the thrust coefficients and anti-torque coefficients of the front and rear rotors, and calculate the total demand control torque based on the pseudo-inverse matrix of the thrust distribution matrix to obtain the basic thrust command of the front and rear rotors. Based on the cross-coupled aerodynamic interference model of the front and rear rotors, the basic thrust command is corrected by anti-torque compensation to generate thrust distribution control command.

[0058] The pseudo-inverse formula for the thrust distribution matrix is ​​as follows:

[0059]

[0060] in, This represents the total demand control torque vector, which includes torque demands in three dimensions: roll, pitch, and yaw. This represents the thrust distribution matrix, which is composed of the geometric lever arms and aerodynamic efficiency of the front and rear rotors. It represents its Moore-Penrose pseudo-inverse matrix, which is used to find the optimal thrust distribution solution in the least squares sense in overdetermined or underdetermined equation systems; This represents the calculated basic thrust command vectors for the front and rear rotors; This represents the anti-torque compensation correction, used to offset the yaw torque imbalance caused by the difference in rotor speeds. In actual calibration, its compensation coefficient is usually obtained through wind tunnel testing and ranges from 0.05 to 0.15. The final output is... This refers to the final control command issued to the electronic control unit (ESC).

[0061] As an optional embodiment, refer to the appendix Figure 8 For adaptive stability control process and execution constraints: The adaptive stability control process specifically includes: Start; Input disturbance level and trend prediction results; Determine control objective (attitude stabilization / trajectory tracking); Adaptive gain adjustment (PID / LQR / MPC); Thrust distribution and attitude control; Rotor cooperative control and anti-torque compensation; Stability optimization and constraint handling; Output control commands; End.

[0062] During the adaptive gain adjustment process, the system dynamically tunes the stiffness gain Kp, damping gain Kd, integral gain Ki, front and rear rotor ratio, and anti-interference weight λ.

[0063] If any rotor thrust value in the thrust distribution control command exceeds the motor's physical maximum thrust threshold (e.g., PWM duty cycle reaches 95% or more), thrust redistribution is triggered, cutting off the rotor thrust value exceeding the motor's physical maximum thrust threshold to the motor's physical maximum thrust threshold, and calculating the thrust loss. Based on the lever arm ratio of the cross-rotor UAV, the thrust deficit is proportionally distributed to the unsaturated rotor, and the anti-torque compensation coefficient of the other rotor is adjusted simultaneously to update the thrust distribution control command.

[0064] As an optional implementation, degradation handling is provided for sensor data packet loss or timeout: If continuous packet loss is detected in micro-meteorological time-series data or multi-source flight status time-series data, and the packet loss duration exceeds a preset timeout threshold (e.g., 50 milliseconds), the inference of the time-series prediction network is paused to save computing power.

[0065] The extended Kalman filter is invoked, and the flight state and wind field disturbance state at the current moment are recursively estimated based on the state vector and system state transition matrix of the last valid moment before packet loss. The gain parameter matrix of the flight controller is downgraded to a fixed parameter mode, and the feedforward compensation torque is calculated based on the recursively estimated wind field disturbance state to maintain basic flight stability.

[0066] Furthermore, to address computing bottlenecks caused by limited edge computing resources or high loads: when the utilization rate of the flight control main CPU exceeds 85% for three consecutive control cycles, model degradation is triggered. The system automatically switches the Transformer model containing a multi-head self-attention mechanism to a gated recurrent unit (GRU) network with fewer parameters, or directly calls the "wind field characteristics-control gain" three-dimensional lookup table generated offline based on historical wind field data for interpolation calculation, sacrificing a slight amount of prediction accuracy to ensure that the control loop meets the hard real-time requirements of over 100Hz.

[0067] In an alternative embodiment, refer to the appendix Figure 9 Based on flight stability control principles, this system constructs a control architecture that integrates state estimation, feedforward compensation, and feedback control. External gusts and turbulent disturbances act on the UAV body, generating six degrees of freedom forces and torques, causing changes in attitude, angular velocity, position, and vibration response. The onboard sensor network collects these state outputs in real time and inputs them to an extended Kalman filter (EKF) for state reconstruction and multi-source data fusion to eliminate measurement noise. The fused high-precision state data is simultaneously fed into two parallel branches of the controller: in the feedback branch, a feedback control algorithm is executed based on the current state error and the reconstructed gain parameter matrix; in the feedforward branch, disturbance compensation feedforward calculation is performed based on the multi-dimensional gust feature vector and the output of the time-series prediction network. The control quantities output from the two branches are superimposed in the thrust distribution and cooperative control module, and after rotor anti-torque compensation and motor physical constraint processing, are sent down to the front / rear rotor drive mechanism. Changes in the rotational speed of the actuators directly alter the aerodynamic forces on the UAV, offsetting external disturbances and ultimately achieving attitude stability and high-precision trajectory tracking, forming wind-resistant stability control.

[0068] Furthermore, the flight stability control closed loop specifically corresponds to the appendix. Figure 9 As shown: External disturbances (including but not limited to gust disturbances, turbulent disturbances, airflow shearing, and topographic effects) generate disturbance forces. ) and torque ( ) acts on the UAV body; the UAV body outputs its state (attitude) , , angular velocity , , The system measures position (x, y, z), velocity (Vx, Vy, Vz), and vibration response. After state estimation (Kalman filtering / EKF, state reconstruction and fusion), the data is input to the controller. The controller integrates adaptive control algorithms, disturbance compensation feedforward, feedback control (PID / LQR / MPC), and thrust distribution and cooperative control, and outputs to the actuators (front rotor drive, rear rotor drive, and tail rotor actuator). Ultimately, it achieves flight performance (attitude stability, trajectory tracking accuracy, vibration suppression, and wind resistance), and forms a state feedback loop through sensor measurements to correct control deviations in real time.

[0069] As an optional implementation, taking a cross-rotor UAV performing an inspection mission in a city as an example, when the UAV is between two tall buildings, it is highly susceptible to sudden strong wind shear caused by the "narrowing effect." At this time, the miniature wind speed sensor on the nose captures the sudden change in wind speed, while the IMU samples the high-frequency vibration of the aircraft. The FPGA coprocessor completes the STFT transformation to extract the violent fluctuation characteristics of the rotor load. The Transformer model on the ARM main controller quickly predicts that a level 3 (severe) lateral gust disturbance will occur in the future. The system reconstructs the control gain, increases the damping gain of the roll channel, and calculates the feedforward compensation torque. During the thrust distribution phase, considering the surge in load on the windward motor caused by the crosswind, the system performs thrust compensation for the windward motor in advance and performs anti-torque correction for the leeward motor. If the windward motor reaches its physical limit, the system immediately triggers the thrust redistribution mechanism to transfer the lost thrust to other rotors. Through the coordinated execution of the aforementioned data streams and hardware, the attitude deviation angle of the drone is strictly limited within a safe threshold when encountering sudden gusts of wind in narrow channels, enabling highly stable flight in complex urban environments.

[0070] As an optional embodiment, refer to the appendix Figure 1 and Figure 2 It also provides an adaptive stability control system for a cross-rotor UAV based on gust perception and trend prediction, including: The feature extraction and fusion module, in response to acquiring multi-source flight status time-series data and micro-meteorological time-series data of the cross-rotor UAV, performs short-time Fourier transform on the airframe vibration signal and rotor speed signal in the multi-source flight status time-series data to extract rotor load fluctuation features; and performs feature splicing and normalization processing on the rotor load fluctuation features and the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time-series data to generate a multi-dimensional gust feature vector. The trend prediction module inputs the multi-dimensional gust feature vector into a pre-trained time-series prediction network and outputs the wind field disturbance state vector and the corresponding disturbance risk level within a preset time window. The control reconfiguration and allocation module dynamically reconfigures the gain parameter matrix of the flight controller based on the disturbance risk level, and calculates the feedforward compensation torque based on the wind field disturbance state vector. Combining the reconfigured gain parameter matrix and the feedforward compensation torque, it generates control commands for the front and rear rotor thrust distribution of the cross-rotor UAV and outputs them to the motor drive unit.

[0071] This application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to any of the foregoing embodiments.

[0072] This application also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to any of the foregoing embodiments.

[0073] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0074] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. An adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction, characterized in that, Includes the following steps: In response to acquiring multi-source flight status time-series data and micro-meteorological time-series data of the cross-rotor UAV, a short-time Fourier transform is performed on the body vibration signal and rotor speed signal in the multi-source flight status time-series data to extract rotor load fluctuation characteristics. The rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data for feature splicing and normalization to generate a multi-dimensional gust feature vector. The multidimensional gust feature vector is input into a pre-trained time-series prediction network, which outputs the wind field disturbance state vector and the corresponding disturbance risk level within a future preset time window. Based on the disturbance risk level, the gain parameter matrix of the flight controller is dynamically reconstructed, and the feedforward compensation torque is calculated based on the wind field disturbance state vector. By combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV and output to the motor drive unit.

2. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, Perform short-time Fourier transform on the airframe vibration signal and rotor speed signal in the multi-source flight state time series data to extract rotor load fluctuation characteristics, including: The airframe vibration signal and the rotor speed signal are windowed and truncated using a Hanning window of a preset length, and sliding sampling is performed according to a preset overlap rate to obtain multiple local time-frequency segments; Perform a fast Fourier transform on each of the local time-frequency segments to extract the energy spectral density of the corresponding frequency band; Based on the peak amplitude of the energy spectral density in the rotor fundamental frequency and its harmonic frequency bands, the fluctuation variance of the rotor aerodynamic load is calculated, and the fluctuation variance is used as the rotor load fluctuation characteristic.

3. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, The rotor load fluctuation characteristics are combined with the wind speed and wind direction change rates corresponding to the micro-meteorological time series data through feature concatenation and normalization to generate a multi-dimensional gust feature vector, including: Calculate the difference of the micro-meteorological time series data within a continuous sampling period to obtain the wind speed change rate and the wind direction change rate; The rotor load fluctuation characteristics, the wind speed change rate, the wind direction change rate, and the turbulence intensity index are concatenated into a one-dimensional initial feature vector; The mean and standard deviation of each dimension in the one-dimensional initial feature vector are calculated based on the historical wind field dataset. The one-dimensional initial feature vector is then normalized using the Z-score normalization algorithm to output the multi-dimensional gust feature vector.

4. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, The temporal prediction network is a Transformer model incorporating a multi-head self-attention mechanism. The multi-dimensional gust feature vector is input into the pre-trained temporal prediction network, which outputs the wind field disturbance state vector and the corresponding disturbance risk level within a preset future time window, including: The multidimensional gust feature vector is mapped into a query vector, a key vector, and a value vector; The attention weight matrix is ​​obtained by calculating the dot product of the query vector and the key vector, dividing it by the scaling factor, and then processing it through the Softmax function. The spatiotemporal evolution features of the wind field are extracted by weighting and summing the value vector using the attention weight matrix. The spatiotemporal evolution features of the wind field are input into a fully connected classification layer, the wind field disturbance state vector is output, and the disturbance risk level is matched and output based on the disturbance intensity threshold and duration threshold in the wind field disturbance state vector. The disturbance risk levels include five levels: no disturbance, slight, moderate, severe, and extremely severe. Based on the disturbance risk level, the gain parameter matrix of the flight controller is dynamically reconstructed, including: When the disturbance risk level is no disturbance or slight disturbance, the inference process of the time series prediction network is bypassed, and the stiffness gain, damping gain and integral gain of the flight controller are determined by calling the preset basic gain parameter table. When the disturbance risk level is medium, the damping gain in the basic gain parameter table is amplified and compensated based on the first weighting coefficient. When the disturbance risk level is severe or extremely severe, the gain adaptive adjustment is triggered. Based on the second weighting coefficient, the stiffness gain and damping gain in the basic gain parameter table are amplified and compensated, and an anti-interference weighting factor is introduced to attenuate the integral gain, thereby generating the reconstructed gain parameter matrix.

5. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, Calculating the feedforward compensation torque based on the wind field disturbance state vector includes: A wind field disturbance model is established based on the wind field disturbance state vector to analyze the three-axis disturbance force and three-axis disturbance torque experienced by the cross-rotor UAV in the body coordinate system. Based on the three-axis disturbance force and the three-axis disturbance torque, the attitude compensation amount and trajectory compensation amount are calculated respectively. The trajectory compensation amount is converted into a desired attitude command through the outer loop position controller, and the attitude compensation amount is converted into a desired angular velocity through the inner loop attitude controller; Based on the desired attitude command and the desired angular velocity, inverse dynamics calculation is performed to generate the feedforward compensation torque.

6. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, Combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV, including: The feedback control torque is calculated based on the current attitude error of the UAV and the reconstructed gain parameter matrix. The feedback control torque is superimposed with the feedforward compensation torque to obtain the total demand control torque; Construct a thrust distribution matrix that includes the thrust coefficients and anti-torque coefficients of the front and rear rotors, and calculate the total demand control torque based on the pseudo-inverse matrix of the thrust distribution matrix to obtain the basic thrust command of the front and rear rotors. Based on the cross-coupled aerodynamic interference model of the front and rear rotors, the basic thrust command is corrected by anti-torque compensation to generate the thrust distribution control command.

7. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, It also includes exception handling logic for motor output saturation: If any rotor thrust value in the thrust distribution control command exceeds the maximum physical thrust threshold of the motor, the thrust redistribution mechanism is triggered. The rotor thrust value exceeding the maximum physical thrust threshold of the motor is truncated to the maximum physical thrust threshold of the motor, and the thrust loss is calculated. Based on the lever arm ratio of the cross-rotor UAV, the thrust deficit is proportionally allocated to the unsaturated rotor, and the anti-torque compensation coefficient of the other rotor is adjusted simultaneously to update the thrust distribution control command.

8. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, It also includes degradation measures for sensor data packet loss or timeouts: If continuous packet loss is detected in the micro-meteorological time-series data or the multi-source flight status time-series data, and the packet loss duration exceeds a preset timeout threshold, the inference of the time-series prediction network is suspended. The extended Kalman filter is invoked, and the flight state and wind field disturbance state at the current moment are recursively estimated based on the state vector and system state transition matrix of the last valid moment before packet loss. The gain parameter matrix of the flight controller is downgraded to a fixed parameter mode, and the feedforward compensation torque is calculated based on the recursively estimated wind field disturbance state.

9. The adaptive stabilization control method for a cross-rotor UAV based on gust perception and trend prediction according to claim 1, characterized in that, It also includes degradation handling for edge computing resource constraints: If the utilization rate of the flight control main processor executing the method exceeds a preset utilization rate threshold for multiple consecutive control cycles, model degradation is triggered. The time series prediction network is switched to a gated recurrent unit network with fewer parameters than the time series prediction network, or a three-dimensional lookup table of wind field characteristics and control gain generated offline based on historical wind field data is called to perform interpolation calculations to update the gain parameter matrix.

10. An adaptive stability control system for a cross-rotor unmanned aerial vehicle based on gust perception and trend prediction, employing the method as described in any one of claims 1 to 9, characterized in that, include: The feature extraction and fusion module is configured to, in response to acquiring multi-source flight state time-series data and micro-meteorological time-series data of the cross-rotor UAV, perform short-time Fourier transform on the body vibration signal and rotor speed signal in the multi-source flight state time-series data to extract rotor load fluctuation features. The rotor load fluctuation characteristics are combined with the wind speed change rate and wind direction change rate corresponding to the micro-meteorological time series data for feature splicing and normalization to generate a multi-dimensional gust feature vector. The trend prediction module is configured to input the multi-dimensional gust feature vector into a pre-trained time-series prediction network and output the wind field disturbance state vector and the corresponding disturbance risk level within a future preset time window. The control reconfiguration and allocation module is configured to dynamically reconfigure the gain parameter matrix of the flight controller based on the disturbance risk level, and calculate the feedforward compensation torque based on the wind field disturbance state vector. By combining the reconstructed gain parameter matrix and the feedforward compensation torque, a forward and rear rotor thrust distribution control command is generated for the cross-rotor UAV and output to the motor drive unit.

Images