A method and system for persistent lock of an unmanned aerial vehicle on a ground target

Abstract

The application discloses a method and system for continuously locking the posture of an unmanned aerial vehicle (UAV) on a target, which comprises the following steps: obtaining the target position and the UAV posture data stream, performing coordinate conversion to generate a target posture relationship identifier, and performing deviation vector decomposition to form a visual axis deviation feature set; performing posture mapping processing to disassemble the visual axis deviation feature set into fast and slow response chains, repairing broken nodes to construct a posture correction trajectory, extracting a control feature mode to generate a response feature factor, and constructing a posture compensation constraint matrix; performing frequency domain decomposition and cross-coupling calculation on the visual axis deviation feature set to determine a coupling correlation degree, identifying a superimposed disturbance sequence, and dynamically reconstructing to generate a compensation gain sequence; performing stability evaluation to determine a time sequence distribution, identifying an abnormal disturbance feature, and extracting a key time control feature; fusing the constraint matrix and the gain sequence to generate a fusion control law, determining a dynamic priority control coefficient, and outputting a target continuous locking posture control instruction, so that the moving target is continuously and stably locked.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle control, in particular to a method and system for attitude control of continuous target locking of unmanned aerial vehicle. BACKGROUND

[0002] When performing tasks such as reconnaissance and surveillance, target tracking, etc., the unmanned aerial vehicle needs to achieve continuous and stable visual locking of ground or air moving targets. In a complex environment, the unmanned aerial vehicle itself is affected by factors such as airflow disturbance and body vibration, and the target itself may also make rapid maneuvers, which makes the visual axis stability control face serious challenges. How to maintain the accurate alignment of the photoelectric load visual axis and the target in a dynamic disturbance environment has become a key technical bottleneck restricting the performance of unmanned aerial vehicle target tracking.

[0003] The existing unmanned aerial vehicle target tracking control method mainly relies on a single feedback control strategy, and when dealing with multiple disturbance sources, it often uses fixed control parameters and unified response modes. This kind of method is difficult to adapt to the time-varying nature of disturbance characteristics, and when multiple disturbance sources act at the same time or the disturbance intensity changes rapidly, the control system is prone to response lag or over-regulation problems. In addition, the traditional method lacks in-depth analysis of the coupling effect between disturbances, and cannot identify the collaborative influence law of different disturbance sources, resulting in a mismatch between the compensation strategy and the actual disturbance characteristics. At the control execution level, the existing method usually uses static gain configuration and a single priority strategy, which is difficult to adaptively optimize according to the dynamic characteristics of the attitude adjustment process, affecting the accuracy and stability of target locking. SUMMARY

[0004] The present application discloses a method and system for attitude control of continuous target locking of unmanned aerial vehicle, aiming to establish an accurate correlation between target space position and unmanned aerial vehicle attitude adjustment through multi-dimensional deviation feature extraction and attitude mapping processing; reveal the collaborative action law of multiple disturbance sources through coupling analysis and superposition processing of disturbance behavior; realize dynamic optimization and adaptive adjustment of control parameters through stability evaluation and abnormality identification; and finally form a control instruction generation mechanism that integrates multi-level information, providing accurate and stable attitude control capability for continuous target locking of unmanned aerial vehicle in a complex disturbance environment.

[0005] The present application discloses a method and system for attitude control of continuous target locking of unmanned aerial vehicle, aiming to establish an accurate correlation between target space position and unmanned aerial vehicle attitude adjustment through multi-dimensional deviation feature extraction and attitude mapping processing; reveal the collaborative action law of multiple disturbance sources through coupling analysis and superposition processing of disturbance behavior; realize dynamic optimization and adaptive adjustment of control parameters through stability evaluation and abnormality identification; and finally form a control instruction generation mechanism that integrates multi-level information, providing accurate and stable attitude control capability for continuous target locking of unmanned aerial vehicle in a complex disturbance environment.

[0006] Obtain the position data stream of the ground target and the attitude data stream of the unmanned aerial vehicle, perform coordinate conversion on the position data stream and the attitude data stream to generate a target attitude relationship identifier, and perform deviation vector decomposition on the target attitude relationship identifier to generate a visual axis deviation feature set;

[0007] The posture mapping processing is performed through the sight axis deviation feature set to identify a posture correction track, a control feature mode is extracted from the posture correction track to perform adaptive association to generate a response feature factor, and a posture compensation constraint matrix is constructed based on the response feature factor;

[0008] A disturbance behavior analysis is performed on the sight axis deviation feature set to determine coupling correlation degrees of different disturbances, the coupling correlation degrees are used to identify a superimposed disturbance sequence, and the posture compensation constraint matrix is dynamically reconstructed according to the superimposed disturbance sequence to generate a compensation gain sequence;

[0009] A stability evaluation is performed on the compensation gain sequence to determine a time sequence distribution of posture adjustment, an abnormal disturbance feature is identified from the time sequence distribution to generate an abnormal disturbance label, and a control feature graph is obtained by extracting a key time control feature from the time sequence distribution;

[0010] The posture compensation constraint matrix and the compensation gain sequence are interval mapped and fused to generate a fusion control law, a dynamic priority control coefficient is determined based on the fusion control law and the abnormal disturbance label, and a target continuous locking posture control instruction is output by using the dynamic priority control coefficient in combination with the control feature graph.

[0011] The second aspect of the present application proposes a posture control system for continuous locking of an unmanned aerial vehicle on a ground target, comprising:

[0012] A data acquisition module is configured to acquire a position data stream of a ground target and an attitude data stream of an unmanned aerial vehicle, perform coordinate conversion on the position data stream and the attitude data stream to generate a target attitude relationship identifier, and perform deviation vector decomposition on the target attitude relationship identifier to generate a sight axis deviation feature set;

[0013] A track identification module is configured to perform posture mapping processing through the sight axis deviation feature set to identify a posture correction track, extract a control feature mode from the posture correction track to perform adaptive association to generate a response feature factor, and construct a posture compensation constraint matrix based on the response feature factor;

[0014] A disturbance analysis module is configured to perform a disturbance behavior analysis on the sight axis deviation feature set to determine coupling correlation degrees of different disturbances, use the coupling correlation degrees to identify a superimposed disturbance sequence, and dynamically reconstruct the posture compensation constraint matrix according to the superimposed disturbance sequence to generate a compensation gain sequence;

[0015] A stability evaluation module is configured to perform a stability evaluation on the compensation gain sequence to determine a time sequence distribution of posture adjustment, identify an abnormal disturbance feature from the time sequence distribution to generate an abnormal disturbance label, and obtain a control feature graph by extracting a key time control feature from the time sequence distribution;

[0016] An instruction generation module is configured to perform interval mapping fusion of the attitude compensation constraint matrix and the compensation gain sequence to generate a fusion control law, determine a dynamic priority control coefficient based on the fusion control law and the abnormal disturbance label, and output a target continuous locking attitude control instruction by combining the dynamic priority control coefficient with the control feature map.

[0017] The beneficial effects of the present application are embodied in the following points: 1. By projecting and analyzing the target attitude relationship label on the shaft system and determining the deviation dominant axis, a priority selection mechanism for attitude adjustment is established, avoiding resource waste and mutual interference when multiple degrees of freedom are adjusted at the same time; further, by separating the fast response chain and the slow response chain, the layered response to sudden deviation changes and long-term tracking trends is realized, combining the identification and repair of the broken node of the trajectory connection label to solve the discontinuity problem of the connection when the double response chains are fused; and by extracting the control feature mode to construct the response relationship graph, the weak coupling collaborative superposition factor is used to quantify the comprehensive applicability of multiple standard scenes to the current scene, so that the attitude compensation can obtain adaptive adjustment basis from multiple source control experiences. 2. By performing frequency domain decomposition and cross-coupling calculation on the boresight deviation feature set, the cooperative influence relationship of disturbances in different frequency bands on the azimuth and elevation angles is identified, solving the compensation mismatch problem caused by the simple superposition of multiple source disturbances and the neglect of phase relationship in traditional methods; by using the coupling correlation degree to generate a superimposed disturbance sequence, the disturbance representation is more consistent with the actual multi-source cooperative action law; and by analyzing and positioning the strength aggregation domain, identifying the jump section of the disturbance and performing gradient-driven reconstruction, the dynamic optimization of the compensation gain according to the time-varying characteristics of the disturbance strength is realized, improving the response lag or overcompensation problem of the fixed gain configuration when the disturbance changes dramatically. 3. By performing multi-layer mapping conversion on the time series distribution to extract extreme value features on short, medium and long time scales, an abnormality identification framework across time scales is established, solving the problem that single time scale analysis easily misses gradual abnormality or misjudges instantaneous fluctuation; by generating an abnormal disturbance label containing double information of abnormality detection and compensation correction, a complete archive of historical abnormal patterns is provided for the control system; when generating the attitude control instruction, the dynamic priority control coefficient is introduced and the corresponding control strategy template is selected by combining the control feature map, realizing the double adaptive adjustment of the control response strength according to the historical abnormal characteristics and the current control mode, avoiding the repeated occurrence of known abnormal patterns, and improving the stability and reliability of the target locking process. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are part of the specification, illustrate specific examples of the technical solutions described in the present application, and constitute a part of the specification, used to explain the technical solutions, principles and effects of the present application.

[0019] Unless specifically stated or otherwise as clear from the context of the foregoing disclosure, the same reference numerals in different drawings can refer to the same or similar techniques. For the same reason, different reference numerals can be meant to refer to the same or similar techniques.

[0020] Figure 1 is a flow diagram of a posture control method for a UAV to continuously lock a target on the ground.

[0021] Figure 2 is a structural block diagram of a posture control system for a UAV to continuously lock a target on the ground. DETAILED DESCRIPTION

[0022] In the following description, for purposes of explanation and not limitation, specific details are set forth, such as a particular sequence of steps, in order to provide a thorough understanding of the present embodiments. However, it will be apparent to one skilled in the art that the present embodiments can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, and circuits are omitted so as not to obscure the description of the present embodiments.

[0023] It is to be understood that the terminology "including", "comprising", "consisting" or "consisting essentially of" used in the specification and the appended claims, indicates open ended inclusion using "one or more" of the recited elements, steps, operations, components, members or a combination thereof, but does not exclude additional elements, steps, operations, components, members or a combination thereof.

[0024] Reference throughout this specification to "one embodiment" or "an embodiment" or "some embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrases "in one embodiment" or "in some embodiments" or "in other embodiments" or "in still other embodiments" or similar phrases in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0025] The technical solutions of the embodiments of the present application are described below.

[0026] As shown in Figure 1 The embodiment of the present application provides a posture control method for a UAV to continuously lock a target on the ground, which comprises the following steps S110-S150:

[0027] In step S110, the position data stream of the ground target and the attitude data stream of the UAV are acquired, the position data stream and the attitude data stream are subjected to coordinate conversion to generate a target attitude relationship identifier, and the target attitude relationship identifier is subjected to bias vector decomposition to generate a boresight deviation feature set.

[0028] Specifically, the position data stream of the ground target and the attitude data stream of the UAV are acquired. The position data stream of the ground target is acquired by an optoelectronic pod carried by the UAV, the optoelectronic pod integrates a high-resolution visible light camera and an infrared thermal imaging camera, and combines a GPS positioning module and a laser range finder to realize real-time measurement of the target position. The position data stream includes three coordinate components of longitude, latitude and altitude of the target, adopts WGS-84 geodetic coordinate system as a reference datum, and the data acquisition frequency is set to 50 Hz to ensure the continuous tracking capability for fast moving targets. The measurement accuracy of the laser range finder reaches ±0.5 meters, and the effective ranging range covers 500 meters to 5000 meters, and the positioning error of the ground vehicle target is controlled within 1.2 meters when cruising at a height of 3000 meters. The position data stream values and timestamp information at each sampling time are recorded to form time series data containing the target motion trajectory. The attitude data stream of the UAV is acquired by an inertial measurement unit (IMU) and a magnetic compass of the UAV, the IMU is composed of a six-axis gyroscope and an accelerometer, and measures three attitude angles of the UAV, i.e., pitch angle, roll angle and yaw angle. The sampling frequency of the attitude data stream is set to 100 Hz, the attitude angle measurement accuracy is ±0.1 degrees, and the attitude stability deviation is not more than ±0.3 degrees under the meteorological condition of wind speed of 8 meters / second. The attitude data stream values and timestamp at each sampling time are recorded to form a continuous data sequence reflecting the flight attitude change of the UAV.

[0029] The target attitude relationship identifier is generated by coordinate conversion of the position data stream and the attitude data stream. The relative position vector of the target relative to the UAV is calculated by extracting the geodetic coordinates of the target in the position data stream and the geodetic coordinates of the UAV in the attitude data stream. The geodetic coordinates in the position data stream are converted into the body coordinate system with the UAV as the origin by using the conversion method from the spherical coordinate system to the Cartesian coordinate system, and the relative position vector contains three components in the east direction, the north direction and the vertical direction. The pitch angle, the roll angle and the yaw angle of the UAV are extracted from the attitude data stream, and the attitude rotation matrix is constructed to convert the relative position vector from the geodetic coordinate system to the UAV body coordinate system. The X-axis of the body coordinate system points to the head direction, the Y-axis points to the right side of the fuselage, and the Z-axis is vertically downward. The position components x_b, y_b and z_b of the target in the body coordinate system are obtained by fusing and calculating the position data stream and the attitude data stream through the rotation matrix, and the dimensions of the three position components are meters. The azimuth angle α and the pitch angle β of the target relative to the body coordinate system are calculated. The azimuth angle is calculated by the inverse tangent function of the ratio of the horizontal component to the forward component, and the pitch angle is calculated by the inverse tangent function of the ratio of the vertical component to the horizontal distance. The value range of the azimuth angle α is-180 degrees to 180 degrees, and the positive value indicates that the target is located on the right side of the fuselage, and the negative value indicates that the target is located on the left side. The value range of the pitch angle β is-90 degrees to 90 degrees, and the positive value indicates that the target is located below the horizontal plane, and the negative value indicates that the target is located above the horizontal plane. The target distance d is calculated by the square root of the sum of squares of the three-axis position components. The azimuth angle α, the pitch angle β and the target distance d of the target are combined to form the target attitude relationship identifier, which completely describes the spatial relationship of the target relative to the current attitude of the UAV.

[0030] In some embodiments, the bias vector decomposition of the target attitude relationship identifier generates an optical axis deviation feature set, including: axis system projection analysis of the target attitude relationship identifier to obtain a three-axis deviation component set; performing amplitude scale calibration on the three-axis deviation component set to determine a dominant axis direction of deviation; converting the dominant axis direction of deviation into an attitude adjustment reference point to form a reference point guide table; and establishing an optical axis deviation feature set according to the reference point guide table.

[0031] The three-axis deviation component set is obtained by projecting and analyzing the target attitude relationship identifier. The azimuth angle a, the pitch angle β and the target distance d in the target attitude relationship identifier are extracted, which describe the spatial position of the target in the field of view of the UAV. The UAV body coordinate system is established, the X-axis points to the head direction, the Y-axis points to the right side of the fuselage, and the Z-axis is perpendicular downward, forming a right-hand coordinate system. The line-of-sight direction vector in the target attitude relationship identifier is projected onto the three axes of the body coordinate system to obtain the deviation components in the three directions. Based on the parameters in the target attitude relationship identifier, the X-axis deviation component Δx is calculated by the formula Δx = d × cos(β) × cos(a), which represents the distance deviation of the target in the head direction, and the dimension is meter. The Y-axis deviation component Δy is calculated by the formula Δy = d × cos(β) × sin(a), which represents the distance deviation of the target in the lateral direction, and the dimension is meter. The Z-axis deviation component Δz is calculated by the formula Δz = d × sin(β), which represents the distance deviation of the target in the vertical direction, and the dimension is meter. The square sum of the three deviation components is equal to the square of the target distance in the target attitude relationship identifier, which satisfies the geometric relationship of vector decomposition. For the target displayed in the target attitude relationship identifier, which is located 15 degrees right of the front of the UAV, 10 degrees in pitch angle and 2000 meters away, the X-axis deviation component is about 1932 meters, the Y-axis deviation component is about 518 meters, and the Z-axis deviation component is about 347 meters. The three deviation components are combined to form a three-axis deviation component set, which completely describes the three-dimensional spatial deviation of the target relative to the UAV attitude reference.

[0032] The amplitude scale calibration is performed on the three-axis deviation component set to determine the deviation dominant axis. Based on the three-axis deviation component set, the absolute values of the deviation components in each axis are calculated, which reflect the amplitude of the deviation in each axis. The three deviation amplitudes in the three-axis deviation component set are normalized by dividing each axis deviation component by the square root of the square sum of the three components. The normalized values are dimensionless and the square sum of the three values is equal to 1, eliminating the influence of distance and making the deviations in each axis comparable. The values of the three normalized deviation components in the three-axis deviation component set are compared, and the axis with the largest value is identified as the deviation dominant axis. When the X-axis normalized deviation is the largest, the X-axis is the deviation dominant axis, indicating that the front-back direction deviation is dominant. When the Y-axis normalized deviation is the largest, the Y-axis is the deviation dominant axis, indicating that the left-right direction deviation is dominant. When the Z-axis normalized deviation is the largest, the Z-axis is the deviation dominant axis, indicating that the vertical direction deviation is dominant. The determination of the deviation dominant axis provides a priority control direction for the UAV attitude adjustment, ensuring that the adjustment action can most effectively reduce the boresight deviation.

[0033] The bias dominant axis is converted into a reference point guide table. According to the type of the bias dominant axis, the priority direction and the adjustment range of the UAV attitude adjustment are determined. When the X-axis is the bias dominant axis, the pitch angle of the UAV is adjusted preferentially, and the reference point is set as the target current pitch angle as the reference value. When the Y-axis is the bias dominant axis, the yaw angle of the UAV is adjusted preferentially, and the reference point is set as the target current azimuth angle as the reference value. When the Z-axis is the bias dominant axis, the flight height of the UAV is adjusted preferentially, and the reference point is set as the height difference between the target and the UAV as the reference value. The target expected value of each reference point is calculated, and the expected value corresponds to the ideal attitude parameter when the line of sight is completely aligned with the target. The reference point guide table is constructed, which includes four fields of the current reference point value, the target expected value, the adjustment direction and the recommended adjustment amount. For the case of X-axis dominance, the reference point guide table records the current pitch angle, the target expected pitch angle, the adjustment direction ("head up" or "head down") and the recommended adjustment amount. The recommended adjustment amount is calculated by the difference between the target expected value and the current value, and the dimension is degree. In the foregoing example, the current pitch angle is -10 degrees, the target expected pitch angle is -9.9 degrees, the adjustment direction is "head up", and the recommended adjustment amount is 0.1 degree. The reference point guide table provides a clear adjustment target and adjustment strategy for the UAV attitude control.

[0034] The line-of-sight bias feature set is established according to the reference point guide table. The bias information of each adjustment reference point is extracted from the reference point guide table, including the angle bias and the adjustment priority. The angle bias is subdivided into azimuth angle bias Δα and pitch angle bias Δβ. The azimuth angle bias Δα is the difference between the target azimuth angle and the expected azimuth angle in the reference point guide table, and the dimension is degree. The pitch angle bias Δβ is the difference between the target pitch angle and the expected pitch angle in the reference point guide table, and the dimension is degree. The change rate of the bias is calculated. The azimuth angle bias change rate dΔα / dt is calculated by the difference between the adjacent time points, and the dimension is degree / second. The pitch angle bias change rate dΔβ / dt is also calculated by the difference, and the dimension is degree / second. The bias convergence index κ is introduced to evaluate the tracking state, and the formula is wherein Δα(t) and Δβ(t) are the azimuth angle deviation and the elevation angle deviation at the current time, Δα(t-Δt) and Δβ(t-Δt) are the azimuth angle deviation and the elevation angle deviation at the last time, Δt is the time interval, and the deviation convergence index κ is a dimensionless value. κ<1 indicates that the deviation is converging, and the line of sight is gradually aligned with the target. κ>1 indicates that the deviation is diverging, and the line of sight deviates from the target. κ=1 indicates that the deviation remains stable. The angle deviations Δα and Δβ, the deviation change rates dΔα / dt and dΔβ / dt, the convergence index κ, and the deviation dominant axis information in the reference point guide table are combined to form a line of sight deviation feature set, and the data structure of the feature set includes six elements. In the target tracking process, the feature values of the line of sight deviation feature set are updated in real time. When the azimuth angle deviation is reduced from 1.5 degrees to 0.3 degrees, the deviation change rate is -0.24 degrees / second, and the convergence index is 0.2, the line of sight deviation feature set shows that the tracking state is good, and the line of sight is quickly aligned with the target.

[0035] In step S120, a posture mapping process is performed on the line of sight deviation feature set to identify a posture correction trajectory, a control feature pattern is extracted from the posture correction trajectory, a response feature factor is generated by adaptive correlation, and a posture compensation constraint matrix is constructed based on the response feature factor.

[0036] In some embodiments, the posture mapping process performed on the line of sight deviation feature set to identify a posture correction trajectory includes: decomposing the line of sight deviation feature set into a fast response chain and a slow response chain; sending a trajectory connection marker from the fast response chain to the slow response chain; collecting the breaking node position information of the trajectory connection marker to form a breaking node list; and selecting the path with the highest reconnection probability at the breaking point as the posture correction trajectory based on the breaking node list.

[0037] The boresight deviation feature set is decomposed into a fast response chain and a slow response chain. The time scale characteristics of each feature parameter in the boresight deviation feature set are analyzed, and the fast-changing features with a response time constant less than 0.5 seconds are classified into the fast response chain, and the slow-changing features with a response time constant greater than 2 seconds are classified into the slow response chain. The fast response chain mainly includes the deviation change rate, convergence index and other feature parameters reflecting the dynamic change of the deviation in the boresight deviation feature set, which can quickly reflect the instantaneous change of the target motion state and the high-frequency disturbance of the unmanned aerial vehicle attitude. The sampling period of the fast response chain is set to 20 milliseconds, which is synchronized with the response period of the unmanned aerial vehicle attitude control, ensuring that the fast-changing deviation can be responded in time. The slow response chain mainly includes the low-frequency component of the angle deviation and the evolution trend of the deviation dominant axis in the boresight deviation feature set, which reflects the long-term change law of the deviation. The sampling period of the slow response chain is set to 200 milliseconds, which is used to capture the overall trend of the target motion and the baseline adjustment requirement of the unmanned aerial vehicle attitude. Through the fast and slow separation processing method, the fast response chain is responsible for sudden deviation changes, and the slow response chain is responsible for maintaining long-term tracking stability. The two response chains are processed and then fused, which can ensure fast response while avoiding the influence of high-frequency noise on long-term tracking accuracy. For a ground vehicle target that suddenly changes direction, the fast response chain quickly captures the jump of the azimuth angle deviation and triggers a rapid adjustment, while the slow response chain provides a stable tracking baseline according to the average motion speed of the target.

[0038] The trajectory connection marker is sent from the fast response chain to the slow response chain. An information transmission mechanism is established between the fast response chain and the slow response chain. When the fast response chain detects a significant deviation mutation, a trajectory connection marker is generated and transmitted to the slow response chain. The trajectory connection marker contains information such as the time of mutation occurrence, mutation amplitude, mutation direction and duration. The mutation amplitude is calculated by the difference between adjacent sampling points, and when the difference value exceeds 3 times the standard deviation of the preset threshold, it is determined as a significant mutation. The fast response chain monitors the deviation change rate at a period of 20 milliseconds, and once the absolute value of the deviation change rate exceeds 5 degrees per second, a trajectory connection marker is immediately generated. The transmission of the trajectory connection marker uses a message queue mechanism to ensure synchronization of information between the fast response chain and the slow response chain. After receiving the trajectory connection marker, the slow response chain adjusts the current long-term tracking baseline and integrates the mutation information captured by the fast response chain into the long-term trend prediction. When the unmanned aerial vehicle encounters a sudden attitude shift due to wind disturbance during tracking, the fast response chain detects that the pitch angle deviation jumps from 0.3 degrees to 1.8 degrees within 100 milliseconds, generates a trajectory connection marker and sends it to the slow response chain, and the slow response chain adjusts the baseline point of the long-term tracking reference trajectory accordingly. The trajectory connection marker establishes a dynamic association between the two response chains, enabling the attitude correction process to respond quickly to mutations while maintaining long-term stability.

[0039] Illustratively, the collection of the broken node position information of the trajectory connection mark forms a broken node list, including: analyzing the connection broken critical point in the trajectory connection mark; performing broken mode recognition on the connection broken critical point to generate a broken feature set; extracting a broken period attribute by means of the broken feature set to obtain a period configuration matrix; and based on the period configuration matrix, performing broken mark labeling on the connection broken critical point to generate a broken node list.

[0040] The connection broken critical point in the trajectory connection mark is analyzed. Each trajectory connection mark is analyzed in detail to extract information such as time stamp, deviation mutation amplitude and transmission state. The transmission delay of the trajectory connection mark is compared with a preset critical delay threshold, which is set to 100 milliseconds. When the transmission delay approaches but does not exceed the critical threshold, the time point corresponding to the trajectory connection mark is marked as a connection broken critical point. The connection broken critical point is located at the boundary position between normal connection and broken connection, and is a warning signal for the deterioration of connection quality. The transmission delay change trend of the trajectory connection mark before and after the connection broken critical point is analyzed, and the time delay rising rate is identified. The time delay rising rate is calculated by dividing the time delay difference of adjacent marks by the time interval, and the dimension is millisecond / second. When the time delay rising rate exceeds 50 milliseconds / second, it indicates that the connection quality is deteriorating rapidly, and the critical point has a high risk of breaking. The position information of all connection broken critical points is recorded, including time stamp, time delay value, time delay rising rate and deviation state before and after the critical point. When the unmanned aerial vehicle performs a maneuvering turn, the transmission delay of the trajectory connection mark gradually rises from an average of 30 milliseconds to 95 milliseconds, which is close to the critical state although it does not exceed the critical threshold of 100 milliseconds. The time point is identified as a connection broken critical point.

[0041] The fracture mode recognition is performed on the connection fracture critical points to generate a fracture feature set. The connection fracture critical points are classified by mode, and the fracture modes are divided into three categories according to the causes and forms of fracture. The first category is sudden fracture, which is characterized by a sharp increase in transmission delay in a short time, usually caused by sudden changes in the attitude of the unmanned aerial vehicle or rapid target maneuver. The second category is cumulative fracture, which is characterized by a gradual increase in transmission delay and eventually exceeding the critical threshold, usually caused by the accumulation of computational load. The third category is periodic fracture, which is characterized by periodic fluctuations in transmission delay and periodically approaching or exceeding the critical threshold, usually caused by periodic disturbance sources such as rotor vibration. By analyzing the time delay curve of each connection fracture critical point, the slope, fluctuation amplitude, and periodicity of the curve are extracted. The slope reflects the speed of the time delay increase, the fluctuation amplitude reflects the instability of the time delay, and the periodicity reflects the regularity of the time delay change. The slope, fluctuation amplitude, periodicity index, and fracture mode type are combined to form the fracture feature set. Each feature parameter in the fracture feature set quantitatively describes the dynamic characteristics of the fracture. For sudden fracture, the fracture feature set shows that the time delay slope is greater than 200 milliseconds per second, the fluctuation amplitude is greater than 50 milliseconds, and there is no obvious periodicity. For periodic fracture, the fracture feature set shows that the time delay slope fluctuates between -50 and 50 milliseconds per second, and the period is about 0.15 seconds corresponding to the vibration period of the rotor speed.

[0042] The fracture period attribute is extracted from the fracture feature set to obtain the period configuration matrix. The fracture critical points with periodic characteristics are selected from the fracture feature set, and the period parameters of these critical points are extracted. The period parameters include period length, period stability, and period amplitude. The period length is obtained by Fourier transform or autocorrelation analysis of the time delay change sequence, and the dimension is seconds. The period stability is obtained by calculating the coefficient of variation of consecutive period lengths, and the smaller the coefficient of variation, the more stable the period. The period amplitude is the peak-to-valley difference of the time delay in the period, and the dimension is milliseconds. The period parameters of each periodic fracture critical point are arranged in chronological order to construct the period configuration matrix. The period configuration matrix is an N×3 matrix, where N is the number of periodic fracture critical points, and the three columns correspond to the period length, period stability, and period amplitude. The period configuration matrix can identify the periodicity of the fracture and determine whether the fracture is caused by a fixed frequency disturbance source. For periodic fracture caused by rotor vibration, the period configuration matrix shows that the period length is stable at about 0.15 seconds, the period stability coefficient of variation is less than 0.05, and the period amplitude fluctuates between 20 and 40 milliseconds.

[0043] The fracture node list is generated by marking the fracture points based on the periodic configuration matrix. According to the periodic information in the periodic configuration matrix, the time points when the fracture may occur in the future are predicted. The prediction method uses periodic extrapolation, assuming that the periodic fracture identified will continue to occur at the same period in the future. For a periodic fracture with a period length of T and a current time of t_0, the fracture may occur at t_0+T, t_0+2T, t_0+3T, etc. The predicted fracture time points are marked as potential fracture nodes, which are included in the fracture node list together with the actual interface fracture critical points identified. Add fracture mark annotations to each fracture node, including fracture type (actual fracture or potential fracture), fracture mode (sudden, cumulative or periodic), fracture severity (transmission failure or only excessive delay) and periodic parameters related to fracture. The fracture severity uses a three-level score, with transmission failure marked as 3 points, delay exceeding 50% of the critical threshold marked as 2 points, and delay exceeding 20% of the critical threshold marked as 1 point. The generated fracture node list arranges all fracture nodes in chronological order, and each node contains a timestamp, a fracture mark annotation and related feature parameters.

[0044] Based on the fracture node list, the path with the highest reconnection probability at the fracture is selected as the attitude correction trajectory. For each fracture node in the fracture node list, the deviation state and adjustment action sequence before and after the fracture are analyzed to find a reconnection path that can reasonably connect the trajectory segments before and after the fracture. The selection of the reconnection path is based on the trajectory continuity principle and the action rationality principle. The trajectory continuity principle requires that the deviation state and adjustment action on the reconnection path remain smooth before and after the fracture, avoiding unreasonable jumps. The action rationality principle requires that the adjustment action corresponding to the reconnection path conforms to the physical constraints and response characteristics of the UAV attitude control. For each fracture node, multiple candidate reconnection paths are generated, each path corresponding to a possible fracture repair scheme. The reconnection probability of each candidate path is calculated, which considers the smoothness of the path, the rationality of the action and the historical success rate. The smoothness of the path is evaluated by calculating the second derivative of the deviation and adjustment action on the reconnection path, and the smaller the second derivative, the smoother the path. The action rationality is evaluated by checking whether the reconnection path meets the rate limit and acceleration limit of the UAV attitude control. The historical success rate is evaluated by statistically analyzing the actual effect of similar reconnection paths in similar fracture situations. The path with the highest reconnection probability is selected as the repair scheme for the fracture node, and the repaired trajectory segment is connected to the overall attitude correction trajectory. All fracture nodes in the fracture node list are processed in sequence to form a complete and continuous attitude correction trajectory, covering the entire process from the initial deviation state to the target alignment state, even in the presence of multiple fracture nodes, the overall coherence and executability of the trajectory can be maintained.

[0045] In some embodiments, the extracting control feature mode from the posture correction trajectory performs adaptive correlation generation of response feature factors, including: constructing a response relationship graph by extracting control feature mode from the posture correction trajectory; obtaining a trajectory matching degree matrix by implementing matching degree measurement on the response relationship graph; obtaining weakly coupled collaborative superposition factors based on weak coupling extraction based on the trajectory matching degree matrix; and using the cumulative intensity value of the weakly coupled collaborative superposition factors as the response feature factors.

[0046] A response relationship graph is constructed by extracting control feature mode from the posture correction trajectory. The deviation state and corresponding adjustment action at each time point on the posture correction trajectory are analyzed to identify the mapping rule between the deviation state and the adjustment action. The posture correction trajectory is segmented according to the deviation dominant axis and the deviation amplitude, and each trajectory segment corresponds to a specific control feature mode. The control feature mode includes parameters such as adjustment direction, adjustment step, adjustment frequency, and adjustment duration. The adjustment direction indicates the increase or decrease of the attitude angle, the adjustment step indicates the angle change amount of a single adjustment, the adjustment frequency indicates the number of adjustments per unit time, and the adjustment duration indicates the time required from the start of adjustment to the convergence of the deviation. The control feature mode parameters of each trajectory segment of the posture correction trajectory are extracted, and a mapping table of deviation state to control feature mode is established. A response relationship graph is constructed, in which the nodes represent different deviation states, the edges represent the transition relationship between the states, and the corresponding control feature mode is marked on the edges. The response relationship graph adopts a directed graph structure, and the direction of the edge represents the evolution direction of the deviation state, which gradually transfers from the initial large deviation state to the small deviation state until the alignment state.

[0047] The matching degree measurement is implemented for the response relationship graph to obtain a trajectory matching degree matrix. Each state transition path in the response relationship graph is compared with a standard trajectory of a historical successful case, and the matching degree of the current trajectory and the standard trajectory is calculated. The matching degree measurement adopts a dynamic time warping algorithm, which can process the stretching and shifting of the trajectory on the time axis. The Euclidean distance between the deviation state of each time point in the current response relationship graph and the corresponding point of the standard trajectory is calculated, and the smaller the distance is, the higher the matching degree is. For the case that the lengths of the trajectories are different, the two trajectories are mapped to a unified time scale through time normalization before being compared. The state transition path of the current response relationship graph is matched with a plurality of standard trajectories respectively, and each standard trajectory corresponds to a typical target tracking scene, such as target uniform straight line motion, target uniform turning, target acceleration motion, etc. A trajectory matching degree matrix is generated, the rows of the matrix correspond to each time point on the current trajectory, the columns correspond to each standard trajectory, and the matrix element value represents the matching degree of the current trajectory at the time point and the corresponding standard trajectory. The matching degree value adopts a normalized score of 0 to 1, 1 represents complete matching, and 0 represents complete mismatch. The trajectory matching degree matrix is an MxK matrix, M is the number of time sampling points of the current trajectory, and K is the number of standard trajectories. Through the trajectory matching degree matrix, it can be identified which standard scene is most similar to the current tracking task, so as to select the corresponding control parameter optimization strategy.

[0048] For example, the weak coupling extraction based on the trajectory matching degree matrix obtains a weak coupling synergistic superposition factor, which includes: developing gradient distribution analysis through the trajectory matching degree matrix to obtain a gradient jump distribution; implementing jump point recognition for the gradient jump distribution to generate a jump point group; extracting a jump amplitude attribute using the jump point group to form an amplitude attribute vector; and performing synergistic superposition amplification based on the amplitude attribute vector to obtain a weak coupling synergistic superposition factor.

[0049] The gradient distribution is obtained by analyzing the trajectory matching degree matrix. The gradient of each column of the trajectory matching degree matrix is calculated, and the gradient represents the rate of change of the matching degree along the time axis. The gradient calculation formula of the element in the ith row and jth column is g_ij=(M_i+1,j-M_i,j) / Δt, where M_i,j is the matrix element value, Δt is the time sampling interval, and the dimension of the gradient g_ij is 1 / s. The positive gradient indicates that the matching degree is rising, and the negative gradient indicates that the matching degree is falling. The gradient jump position in the trajectory matching degree matrix is identified, and the jump refers to a significant mutation of the gradient value between adjacent time points. The second-order difference Δg_i=(g_i+1,j-g_i,j) / Δt of the gradient is calculated, and the second-order difference reflects the change speed of the gradient, with a dimension of 1 / s². When the absolute value of the second-order difference exceeds the threshold value of 0.5 / s², it is determined that the position is a gradient jump point. The gradient jump usually corresponds to the time when the control strategy of the current scene changes or the target motion mode suddenly changes. The positions and jump amplitudes of the gradient jump points in each column of the trajectory matching degree matrix are counted to form the gradient jump distribution. The gradient jump distribution records the time and mutation intensity when each standard trajectory exhibits a matching degree mutation in the current scene. For the scene where the target suddenly turns, the gradient jump distribution of the corresponding column of the uniform straight line motion standard trajectory shows that the matching degree gradient jumps from positive to negative at the start time of the turn, and the jump amplitude reaches 1.2 / s.

[0050] The jump point group is generated by identifying the jump points in the gradient jump distribution. All jump points in the gradient jump distribution are traversed, and classification and screening are performed according to the characteristic parameters of the jump points. The characteristic parameters of the jump point include the jump time, the jump amplitude, the gradient sign before and after the jump, and the jump duration. The jump amplitude is the absolute value of the difference between the gradient values before and after the jump, with a dimension of 1 / s. The jump duration is the time required for the gradient to recover to stability from the start of the jump, with a dimension of seconds. The jump points with a jump amplitude greater than 0.8 / s and a jump duration less than 1 second in the gradient jump distribution are screened out. Such jump points correspond to the rapid switching of the control strategy and are important indicators of weak coupling association. The screened jump points are organized into a jump point group in chronological order, and each element in the jump point group records the complete characteristic information of the jump point. Cluster analysis is performed on the jump point group to identify jump clusters composed of multiple jump points close in time. The jump cluster indicates that multiple standard trajectories simultaneously exhibit matching degree jumps in a short period of time, corresponding to the key turning stage of the current scene. The standard trajectory and jump type (gradient changes from positive to negative, from negative to positive, or the sign remains the same but the amplitude suddenly changes) to which each jump point in the jump point group belongs are labeled.

[0051] The amplitude attribute vector is formed by extracting the amplitude of the jump point group. The jump amplitude value of each jump point in the jump point group is extracted, and the jump amplitude reflects the intensity of the change in the matching degree. The jump amplitudes of the jump points in the jump point group are grouped according to the corresponding standard track numbers, and the average jump amplitude and the maximum jump amplitude corresponding to each standard track are calculated. The average jump amplitude represents the overall weak coupling strength of the standard track in the current scene, and the maximum jump amplitude represents the most significant weak coupling correlation point. The amplitude attribute vector is constructed, the length of the vector is K (the number of standard tracks), and the jth element of the vector is the average jump amplitude of the jth standard track. The amplitude attribute vector has a dimension of 1 / s, and the larger the value of the element in the vector, the stronger the weak coupling relationship between the corresponding standard track and the current scene. The amplitude attribute vector is normalized by dividing each element in the vector by the length of the vector, so that the length of the normalized vector is 1, and the vector becomes dimensionless. The normalized amplitude attribute vector eliminates the influence of the dimension and the numerical scale, and retains the relative relationship between the weak coupling strengths of the standard tracks.

[0052] The weak coupling collaborative superposition factor is obtained based on the amplitude attribute vector. The normalized amplitude attribute vector and the historical success rate vector of each standard track are multiplied element by element to obtain a weighted amplitude vector. The jth element of the historical success rate vector is the probability of successful application of the jth standard track in the historical case, and the value range is 0 to 1. The weighted amplitude vector considers both the weak coupling strength and the historical reliability, and a larger element value indicates that the standard track has both strong weak coupling correlation and high historical success rate. The elements of the weighted amplitude vector are accumulated and summed, and the cumulative sum represents the total amount of collaborative contribution of all weak coupling standard tracks. An amplification coefficient is introduced to adjust the cumulative sum, and the amplification coefficient is determined according to the complexity of the current scene. The higher the scene complexity, the larger the amplification coefficient, to compensate for the uncertainty of weak coupling information in complex scenes. The weak coupling collaborative superposition factor F_wc = α × Σ(V_a) is calculated, where α is the amplification coefficient, the value range is 1.0 to 1.5, V_a is the weighted amplitude vector, and Σ(V_a) represents the sum of the elements of the vector. The weak coupling collaborative superposition factor is a dimensionless value, and the larger the value, the more control strategy references can be obtained from multiple weak coupling standard scenes.

[0053] The cumulative intensity value of the weak coupling synergistic superposition factor is used as a response characteristic factor. The cumulative intensity of the weak coupling synergistic superposition factor in the entire attitude correction trajectory time window is calculated, and the cumulative intensity is obtained by time integration of the weak coupling synergistic superposition factor at each time point. The time integration formula is R = ∫F_wc(t)dt, the integration interval is the time span from the beginning of the trajectory to the end of the trajectory, R is the cumulative intensity value, and F_wc(t) is the weak coupling synergistic superposition factor at time t. Since the weak coupling synergistic superposition factor is a dimensionless value, the dimension of the cumulative intensity value R is dimensionless value multiplied by time. In order to compare the response characteristics of trajectories of different lengths, the cumulative intensity value R is normalized by dividing it by the total length of the trajectory T to obtain the normalized cumulative intensity value R_norm = R / T. The dimension of the normalized cumulative intensity value is dimensionless. The normalized cumulative intensity value is used as a response characteristic factor, which quantifies the comprehensive adaptability of the current attitude correction trajectory to multiple control strategies. A larger response characteristic factor value indicates that the current scenario can flexibly draw on the control experience of multiple standard scenarios, and the attitude compensation control has strong adaptability. A smaller response characteristic factor value indicates that the weak coupling relationship between the current scenario and the standard scenario is weak, and more real-time feedback is needed for control.

[0054] A posture compensation constraint matrix is constructed based on the response characteristic factor. The constraint strength of the posture compensation control is determined according to the value of the response characteristic factor. The larger the response characteristic factor, the stronger the scenario adaptability, and the constraint can be relatively loose to allow greater adjustment flexibility. The smaller the response characteristic factor, the higher the scenario uncertainty, and the constraint needs to be more stringent to ensure control stability. A posture compensation constraint matrix is constructed, which is a 3x3 symmetric matrix with three dimensions corresponding to the pitch angle, yaw angle and flight height three control degrees of freedom. The diagonal elements of the matrix represent the upper limit of the adjustment amplitude of each degree of freedom, and the non-diagonal elements represent the coupling constraint coefficients between different degrees of freedom. The upper limit of the pitch angle adjustment amplitude is determined according to the response characteristic factor and the deviation dominant axis. When the X-axis is the deviation dominant axis and the response characteristic factor is greater than 0.6, the upper limit of the pitch angle adjustment amplitude is set to 5 degrees per second. When the response characteristic factor is less than 0.4, the upper limit of the pitch angle adjustment amplitude is tightened to 2 degrees per second. The upper limits of the adjustment amplitudes of the yaw angle and the flight height are determined in a similar manner. The coupling constraint coefficient reflects the mutual influence when multiple degrees of freedom are adjusted simultaneously. When the pitch angle and the yaw angle are adjusted simultaneously, the coupling constraint coefficient limits the product of the adjustment amplitudes of the two to be less than a certain threshold. The posture compensation constraint matrix provides a clear constraint boundary for the attitude control executor, ensuring that the posture compensation action is performed within a safe and controllable range.

[0055] Step S130, the coupling correlation degrees of different disturbances are determined by performing disturbance behavior analysis on the set of visual axis deviation features, the coupling correlation degrees are used to identify a superimposed disturbance sequence, and a compensation gain sequence is generated by dynamically reconstructing the attitude compensation constraint matrix according to the superimposed disturbance sequence.

[0056] In some embodiments, the coupling correlation degrees of different disturbances are determined by performing disturbance behavior analysis on the set of visual axis deviation features, including: performing frequency domain decomposition on the set of visual axis deviation features to obtain a frequency band amplitude distribution spectrum; performing cross-coupling calculation on the frequency band amplitude distribution spectrum to generate a mutual coupling response channel; extracting response transmission features on the mutual coupling response channel as disturbance coupling features; and summarizing the disturbance coupling features as the coupling correlation degrees of different disturbances.

[0057] The set of visual axis deviation features is subjected to frequency domain decomposition to obtain a frequency band amplitude distribution spectrum. The time series of azimuth angle deviation Δα and pitch angle deviation Δβ in the set of visual axis deviation features are extracted, and fast Fourier transform is performed on the two deviation sequences respectively to convert time domain signals into frequency domain signals. Fast Fourier transform can decompose complex time domain waveforms into sinusoidal components of different frequencies, and each frequency component corresponds to a periodic disturbance. The frequency domain signal contains amplitude information and phase information, the amplitude reflects the energy intensity of the frequency component, and the phase reflects the time offset of the frequency component. The amplitude spectrum of the frequency domain signal is extracted, the horizontal axis is frequency (unit: Hz), and the vertical axis is amplitude (unit: degree). The amplitude spectrum directly shows the energy distribution of the deviations in the set of visual axis deviation features in each frequency band. The frequency axis is divided into multiple frequency band intervals, the low frequency band is 0.1 Hz to 0.5 Hz, the medium frequency band is 0.5 Hz to 5 Hz, and the high frequency band is 5 Hz to 20 Hz. The frequency band amplitude distribution spectrum is generated, and the peak frequency and corresponding amplitude of each frequency band are marked in the spectrum. The frequency band amplitude distribution spectrum of the azimuth angle deviation shows that there is a significant peak at 1.2 Hz, and the amplitude reaches 0.8 degrees, which corresponds to the periodic disturbance caused by the rotation of the rotor. The frequency band amplitude distribution spectrum of the pitch angle deviation shows that there is a peak at 0.3 Hz, and the amplitude reaches 1.2 degrees, which corresponds to the low-frequency fluctuation caused by the air flow disturbance.

[0058] The cross-coupling algorithm is performed on the band amplitude distribution spectrum to generate the cross-coupling response channels. The band amplitude distribution spectrum of the azimuth angle deviation and the pitch angle deviation are cross-compared to identify the bands that have peak values at the same or similar frequencies. The peak values at the same frequency indicate that the disturbance at this frequency simultaneously affects the azimuth angle and the pitch angle, and there is a coupling relationship between the two degrees of freedom. The amplitude product of the two deviation spectra at each frequency point is calculated, and the product value reflects the coupling strength of the two degrees of freedom at this frequency. The frequency point with a larger amplitude product corresponds to a strong coupling band, and the frequency point with a smaller amplitude product corresponds to a weak coupling or uncoupling band. The amplitude product sequence is threshold filtered, and the frequency points with product values exceeding the threshold are selected as the coupling frequency points. The threshold is set to twice the average value of the amplitude product sequence to ensure that significant coupling features are selected. The coupling frequency points and their corresponding amplitude product values are organized into a coupling frequency list, and the list is sorted in descending order of the amplitude product. The cross-coupling response channels are established, and the number of channels is equal to the number of coupling frequency points. Each channel corresponds to a coupling frequency and its coupling strength. The cross-coupling response channels describe the cooperative response characteristics of the azimuth angle and the pitch angle at a specific frequency. At a frequency of 1.2 Hz, the amplitude product of the azimuth angle and the pitch angle reaches 0.96 degrees square, and the first cross-coupling response channel is established. This channel reflects the coupling effect of the rotor vibration on the dual-axis attitude. At a frequency of 0.3 Hz, the amplitude product is 1.44 degrees square, and the second cross-coupling response channel is established. This channel reflects the coupling effect of the airflow disturbance.

[0059] The response transfer feature is extracted from the cross-coupling response channel as the disturbance coupling feature. The time-domain waveform of the frequency component corresponding to each cross-coupling response channel is obtained by time-domain inversion. The time-domain waveform contains amplitude envelope and phase evolution information. The amplitude envelope describes the time-varying characteristics of the coupling strength, and the phase evolution describes the time sequence relationship of the responses of the two degrees of freedom. The phase difference of the azimuth angle and the pitch angle at the frequency component is calculated. The phase difference is determined by the peak position of the cross-correlation function, and the dimension is radian or degree. A phase difference close to 0 degrees indicates that the two degrees of freedom respond synchronously, a phase difference close to 90 degrees indicates that the two degrees of freedom have a quarter-period delay response, and a phase difference close to 180 degrees indicates that the two degrees of freedom respond in opposite directions. The stability of the phase difference is analyzed. A stable phase difference indicates that the coupling mechanism is fixed, and an unstable phase difference indicates that the coupling mechanism changes over time. The response transfer coefficient is calculated. The transfer coefficient is the ratio of the amplitude of the pitch angle deviation to the amplitude of the azimuth angle deviation, and the dimension is a dimensionless value. A transfer coefficient greater than 1 indicates that the disturbance has a greater impact on the pitch direction than the azimuth direction, and a transfer coefficient less than 1 indicates that the azimuth direction has a greater impact. The phase difference, phase stability, and response transfer coefficient are combined to form the response transfer feature, which quantitatively describes the transfer law of the disturbance between the two degrees of freedom. The response transfer feature is the disturbance coupling feature, which reflects the multi-degree-of-freedom coupling influence mechanism of the disturbance. In the 1.2 Hz channel corresponding to the rotor vibration, the phase difference is 45 degrees and the stability variation coefficient is less than 0.1, the response transfer coefficient is 0.75, and the disturbance coupling feature shows that the azimuth angle response is slightly stronger than the pitch angle response and that there is a fixed phase delay relationship between the two.

[0060] The disturbance coupling features of each cross-coupling response channel are summarized and classified according to the type of disturbance source. Channels with similar frequencies and similar transfer features are classified into the same disturbance source category, and channels with different frequencies but significantly different transfer mechanisms are classified into different disturbance source categories. The comprehensive coupling strength of each disturbance source category is calculated. The comprehensive coupling strength is obtained by dividing the sum of the amplitude products of all channels in the category by the number of channels, and the dimension is degree square. The greater the comprehensive coupling strength, the more significant the coupling influence of the disturbance source on the dual-axis attitude. The average phase difference and the average transfer coefficient of each disturbance source category are calculated. The average phase difference reflects the time sequence relationship of the dual-axis response caused by the disturbance source, and the average transfer coefficient reflects the distribution ratio of the influence of the disturbance source in the two directions. The coupling correlation degree is defined as the normalized value of the comprehensive coupling strength. The comprehensive coupling strength of each disturbance source is divided by the sum of the comprehensive coupling strengths of all disturbance sources, so that the sum of the coupling correlation degrees of all disturbance sources is equal to 1. The coupling correlation degree is a dimensionless value, with a value range of 0 to 1. The greater the value, the more important the coupling influence of the disturbance source. The coupling correlation degree of the rotor vibration disturbance is 0.42, the coupling correlation degree of the airflow disturbance is 0.35, the coupling correlation degree of the body structure vibration is 0.18, and the coupling correlation degree of the sensor noise is 0.05.

[0061] The coupling correlation degree is used to identify the superimposed disturbance sequence. According to the coupling correlation degree of each disturbance source, the disturbance source with a coupling correlation degree greater than 0.15 is selected as the main disturbance source. The main disturbance sources include rotor vibration, airflow disturbance and aircraft structure vibration, and the sum of the coupling correlation degrees of the three types of disturbances reaches 0.95, occupying an absolute dominant position. The time-domain waveform of each main disturbance source corresponding to the frequency band is extracted, and the waveforms of different disturbance sources are weighted and superimposed according to their coupling correlation degrees. The superimposition formula is S(t)=∑(C_i×D_i(t)), where S(t) is the superimposed disturbance sequence, C_i is the coupling correlation degree of the i th disturbance source, D_i(t) is the time-domain waveform of the i th disturbance source, and ∑ represents the summation of all main disturbance sources. The phase relationship between the disturbance sources needs to be considered in the superimposition process, and the phase difference is determined by the disturbance coupling characteristics extracted above. When the phase difference of two disturbance sources is close to 0 degrees, the wave peaks are superimposed to produce an enhancement effect, and the amplitude increases after superimposition. When the phase difference is close to 180 degrees, the wave peaks and wave troughs are superimposed to produce a cancellation effect, and the amplitude decreases after superimposition. The generated superimposed disturbance sequence reflects the overall disturbance effect under the cooperative action of multiple main disturbance sources. The amplitude envelope of the superimposed disturbance sequence shows obvious time-varying characteristics. At some time, multiple disturbance sources are in phase and superimposed, resulting in a significant increase in disturbance intensity. At other times, the disturbances of different phases cancel each other out, resulting in a decrease in disturbance intensity.

[0062] In some embodiments, the dynamic reconstruction of the attitude compensation constraint matrix according to the superimposed disturbance sequence to generate a compensation gain sequence includes: performing disturbance aggregation intensity analysis on the superimposed disturbance sequence to locate an intensity aggregation domain; generating a jump mark based on the intensity aggregation domain; performing gradient-driven reconstruction on the jump mark to obtain a potential gain candidate; and determining a compensation gain sequence based on the potential gain candidate and the attitude compensation constraint matrix.

[0063] The disturbance aggregation intensity analysis is performed on the superimposed disturbance sequence to locate an intensity aggregation domain. The superimposed disturbance sequence is analyzed by a sliding window, the window length is set to 2 seconds, and the sliding step is set to 0.5 seconds. The root mean square value of the disturbance amplitude in each window is calculated, and the root mean square value reflects the disturbance energy intensity in the time window. The root mean square value calculation formula is where S(t) is the amplitude of the superimposed disturbance sequence at time t, N is the number of sampling points in the window, and ∑ represents the summation of all sampling points in the window. The dimension of the root mean square value is the same as that of the superimposed disturbance sequence, which is degree. The root mean square values of each window are compared with the global average root mean square value to identify the windows with root mean square values significantly higher than the average level. The intensity aggregation determination threshold is set to 1.5 times the global average root mean square value, and when the window root mean square value exceeds the threshold, it is determined to be an intensity aggregation window. Adjacent intensity aggregation windows are merged to form continuous intensity aggregation domains, which represent the time segments during which the disturbance intensity is continuously higher than the average level. The start time, end time, duration, and average intensity of each intensity aggregation domain are recorded. The average intensity is the arithmetic mean of the root mean square values of all windows in the domain, and the dimension is degree. The duration is the time difference between the end time and the start time, and the dimension is seconds. The distribution characteristics of the intensity aggregation domains are analyzed to identify the frequency and periodicity of the aggregation domains. During the low-speed cruise phase of the UAV, the intensity aggregation domains appear every 15 seconds, with a duration of 3 to 5 seconds and an average intensity of 1.8 degrees, corresponding to the intermittent effects of gust disturbances. During the maneuvering turn phase of the UAV, the intensity aggregation domains continuously appear for up to 12 seconds, with an average intensity of 2.5 degrees, corresponding to the sustained strong disturbance during the maneuvering process.

[0064] Intensity jump section positioning is performed based on the intensity aggregation domains to generate jump markers. The boundary positions of each intensity aggregation domain, including the start point and end point of the aggregation domain, are analyzed in detail. The intensity change rate before and after the boundary position is calculated, which is the difference between the root mean square values of adjacent windows divided by the time interval, with a dimension of degree / second. The intensity change rate at the start point position is usually positive and has a large value, indicating a rapid rise in disturbance intensity. The intensity change rate at the end point position is usually negative and has a large absolute value, indicating a rapid decline in disturbance intensity. The positions where the absolute value of the intensity change rate exceeds the jump threshold are identified, and the jump threshold is set to 1.0 degree / second. When the absolute value of the intensity change rate exceeds the jump threshold, the position is marked as an intensity jump point. The intensity jump point corresponds to the key moment when the disturbance intensity changes abruptly, which is the moment when the compensation control needs to respond quickly. The intensity jump points are arranged in chronological order to form a jump point sequence. The jump direction of the jump points is analyzed, with up-jump points corresponding to an increase in intensity and down-jump points corresponding to a decrease in intensity. The up-jump points and down-jump points are paired to form intensity jump sections, each section starting from an up-jump point and ending at the next down-jump point, corresponding to a complete strong disturbance process. Jump markers are generated for each jump section, including the section number, start time, end time, and maximum intensity.

[0065] The gradient-driven reconstruction is performed on each jump marker to obtain potential gain candidates. The intensity gradient information in each jump marker is extracted. The intensity gradient is the average rate of intensity change in a section, which is calculated by the difference between the maximum intensity and the initial intensity divided by the duration of the section. The unit of the intensity gradient is degree / s. The greater the intensity gradient, the faster the disturbance rises, and a more aggressive compensation response is needed. According to the intensity gradient, the jump section corresponding to each jump marker is divided into three categories: fast-rising section (gradient greater than 1.5 degree / s), medium-rising section (gradient between 0.5 and 1.5 degree / s), and slow-rising section (gradient less than 0.5 degree / s). Different compensation gain adjustment strategies are set for different categories of sections. The fast-rising section adopts an aggressive gain strategy, with the proportional gain increased to 1.8 times the baseline value and the derivative gain increased to 2.0 times the baseline value, to quickly respond to disturbance mutations. The medium-rising section adopts a moderate gain strategy, with the proportional gain increased to 1.3 times the baseline value and the derivative gain increased to 1.5 times the baseline value. The slow-rising section adopts a conservative gain strategy, with the proportional gain maintained at the baseline value and the derivative gain slightly increased to 1.1 times the baseline value. The jump section corresponding to each jump marker generates a corresponding compensation gain configuration scheme, which contains the numerical combination of proportional gain, integral gain, and derivative gain. The gain configuration scheme of each section is taken as a potential gain candidate, and the potential gain candidate set contains multiple gain options for different disturbance jump characteristics.

[0066] The compensation gain sequence is determined based on the potential gain candidate and the attitude compensation constraint matrix. The potential gain candidate is matched with the current attitude compensation constraint matrix to check whether the candidate gain meets the gain upper limit and coupling constraint specified by the constraint matrix. The diagonal elements of the attitude compensation constraint matrix specify the gain upper limit of each degree of freedom, and the non-diagonal elements specify the coupling constraint relationship of the gains of different degrees of freedom. When the candidate proportional gain K_p and the candidate differential gain K_d meet K_p≤K_p_max and K_d≤K_d_max, the gain upper limit test is passed. When the candidate pitch gain K_pitch and the candidate yaw gain K_yaw meet K_pitch×K_yaw≤coupling constraint threshold, the coupling constraint test is passed. The coupling constraint threshold is determined by the non-diagonal elements of the attitude compensation constraint matrix, and has the dimension of the square of the gain unit. The candidates that pass both the gain upper limit test and the coupling constraint test are screened to form a qualified candidate set. The performance of each candidate in the qualified candidate set is evaluated, and the performance evaluation indexes include response speed, stability margin, and energy efficiency. The response speed is estimated by the closed-loop bandwidth corresponding to the candidate gain, and the larger the bandwidth, the faster the response. The stability margin is evaluated by the phase margin and amplitude margin corresponding to the candidate gain, and the larger the margin, the more stable. The energy efficiency is estimated by the control output amplitude corresponding to the candidate gain, and the smaller the amplitude, the lower the energy consumption. The three indexes are combined to score each candidate, and the candidate with the highest score is selected as the optimal compensation gain of the jump section. The optimal compensation gains of each jump section are arranged in chronological order, and transition gains are inserted between the sections to ensure the smoothness of the gain change. The transition gain is generated by linear interpolation between the gains of adjacent sections, and the interpolation duration is set to 0.5 seconds. The complete compensation gain sequence is formed, and the compensation gain sequence covers the entire disturbance action period, so that the compensation control can quickly respond to disturbance mutations and maintain the stability and energy efficiency of the control process.

[0067] In step S140, stability evaluation is performed on the compensation gain sequence to determine the time sequence distribution of attitude adjustment, to identify abnormal disturbance features from the time sequence distribution to generate an abnormal disturbance label, and to extract key time control features from the time sequence distribution to obtain a control feature spectrum.

[0068] Specifically, the stability evaluation of the compensation gain sequence determines the timing distribution of the attitude adjustment. The proportional gain, integral gain and derivative gain values at each time in the compensation gain sequence are extracted, and the gain parameters are substituted into the closed-loop transfer function of the UAV attitude control to calculate the eigenvalue distribution. The real part of the eigenvalue reflects the convergence speed, and when the real part of all eigenvalues is negative, it is stable. The phase margin and amplitude margin corresponding to each gain configuration at each time are calculated, and it is determined to be a stable configuration when the phase margin is greater than 45 degrees and the amplitude margin is greater than 6 dB. The response time of the attitude adjustment under each stable configuration is evaluated, and the response time is the time required from applying the control command to the attitude deviation falling to 10% of the target value, with the dimension being seconds. According to the stability evaluation results, the time axis corresponding to the compensation gain sequence is divided into multiple adjustment periods. The high-gain configuration is used in the fast adjustment period, the response time is controlled within 0.3 seconds, and it is suitable for the disturbance mutation stage. The medium-gain configuration is used in the smooth adjustment period, the response time is controlled within 0.8 seconds, and it is suitable for the disturbance stable stage. The low-gain configuration is used in the fine adjustment period, the response time is controlled within 1.5 seconds, and it is suitable for the end stage close to the target alignment. The start time, end time, gain configuration and stability index of each adjustment period are recorded to form the timing distribution of the attitude adjustment.

[0069] In some embodiments, the identification of the abnormal disturbance feature from the timing distribution generates an abnormal disturbance label, including: implementing multi-layer mapping conversion on the timing distribution to generate a mapping extreme value set; identifying an abnormal disturbance feature through the mapping extreme value set to generate a deviation mark; performing deviation compensation processing based on the deviation mark to generate a compensation disturbance index string; and fusing and arranging the compensation disturbance index string with the mapping extreme value set into an abnormal disturbance label.

[0070] The timing distribution is implemented by multi-layer mapping conversion to generate a mapping extreme value set. Multi-scale analysis is performed on the gain value sequence in the timing distribution to extract extreme value features of gain changes at different time scales. The first layer mapping uses short-time scale analysis, with a time window set to 0.5 seconds and a sliding step set to 0.1 seconds. The local maximum and minimum values of the gain sequence are identified within each window, with the local maximum corresponding to the peak configuration of the gain within the window and the local minimum corresponding to the valley configuration of the gain. The time position and value size of each local extreme value are recorded to form a short-time scale extreme value sequence. The second layer mapping uses medium-time scale analysis, with the time window extended to 2 seconds and the sliding step set to 0.5 seconds. The short-time extreme value sequence is filtered again within each medium-time window to identify the global extreme value within the medium-time window. The third layer mapping uses long-time scale analysis, with the time window extended to 5 seconds to identify the absolute extreme value in the entire compensation process. The extreme value information of the three levels is summarized to form a mapping extreme value set, and each element in the set includes the time position, value size, extreme value type (maximum or minimum) and level (short-time, medium-time or long-time) of the extreme value.

[0071] The abnormal disturbance feature is identified by generating a deviation mark through mapping the extreme value set. The gain value corresponding to each extreme value point in the mapping extreme value set is analyzed, and the deviation degree of the extreme value from other extreme values in the same level is calculated. The deviation degree is calculated by dividing the difference between the extreme value and the average value of the extreme values in the level by the standard deviation to obtain a normalized deviation score, which is a dimensionless value. If the absolute value of the normalized deviation score is greater than 2, it indicates that the extreme value deviates significantly from the normal level, and it is determined as an abnormal extreme value. The abnormal extreme value corresponds to the time when the gain configuration appears abnormal fluctuation, which may be caused by disturbance mutation, fault or improper control parameter configuration. The abnormal extreme value points in each level are identified, and the time sequence distribution state information of the abnormal extreme value corresponding to the time in the mapping extreme value set is extracted. The state information includes the stability index, response time and adjustment phase type at that time. The gain change trend before and after the abnormal extreme value is analyzed to determine whether the abnormality is caused by sudden increase or sudden decrease of the gain. The abnormality corresponding to the sudden increase of the gain is the over-response of the control to the disturbance, which may cause overshoot and oscillation. The abnormality corresponding to the sudden decrease of the gain is the insufficient response of the control, which may cause the deviation to exist continuously and cannot converge. A deviation mark is generated for each abnormal extreme value, and the mark includes the abnormal occurrence time, the deviation direction (up or down), the deviation amplitude and the abnormal type (over-response or insufficient response).

[0072] Based on the deviation mark, a compensation disturbance index string is generated by performing deviation compensation processing. For the abnormal time marked by each deviation mark, a corresponding compensation correction scheme is designed. For the abnormality of excessive increase of gain, the compensation correction scheme adopts the gain reduction strategy, which reduces the gain value in the abnormal time and the subsequent 1 second time period by a certain proportion. The reduction proportion is determined according to the deviation amplitude in the deviation mark. When the deviation amplitude is 2 times the standard deviation, the reduction proportion is set to 20%, and when the deviation amplitude is 3 times the standard deviation, the reduction proportion is set to 35%. For the abnormality of sudden decrease of gain, the compensation correction scheme adopts the gain promotion strategy, which promotes the gain value in the abnormal time and the subsequent time period by a certain proportion. The promotion proportion is also dynamically determined according to the deviation amplitude in the deviation mark. The gain configuration after correction is re-tested for stability to ensure that the corrected configuration still meets the stability requirements. The correction scheme that passes the stability test is adopted, and the scheme that fails the test needs to be adjusted to re-test. A disturbance compensation index is generated for each correction scheme, which includes the start and end time of the correction period, the gain value before and after the correction, the correction type (reduction or promotion) and the correction amplitude. All disturbance compensation indexes are concatenated in time sequence to form a compensation disturbance index string, which records the complete correction operation sequence for abnormal disturbance features and describes how to eliminate the adverse effects caused by abnormalities through gain adjustment.

[0073] The compensation disturbance index string and the mapping extreme value set are fused to arrange an abnormal disturbance label. A corresponding relationship between each index item in the compensation disturbance index string and an abnormal extreme value point in the mapping extreme value set is established, and each index item is associated with an abnormal extreme value point triggering the correction operation. Multi-scale feature information of the abnormal extreme value point is extracted, including the performance of the extreme value at three levels of short time, medium time and long time. The short time level shows the instantaneous mutation characteristics of the abnormality, the medium time level shows the influence range of the abnormality on the surrounding period, and the long time level shows the relative position of the abnormality in the overall compensation process. The extreme value feature information is fused with the corresponding compensation index information to generate a comprehensive label containing double information of abnormality detection and abnormality processing. The structure of the comprehensive label includes six fields of abnormality identification code, occurrence time, extreme value feature description, deviation degree quantization, compensation correction scheme and correction effect evaluation. The abnormality identification code adopts hierarchical coding mode, the first bit represents the abnormality level (1 for short time, 2 for medium time, and 3 for long time), the second bit represents the abnormality type (1 for excessive response and 2 for insufficient response), and the subsequent bits represent the abnormality serial number. All comprehensive labels are sorted and arranged in time sequence and abnormality severity, and the abnormality with high severity has higher priority. The generated abnormal disturbance label set enables the control to recognize historical abnormal patterns and take preventive measures to avoid similar abnormalities from occurring again.

[0074] Control feature maps are obtained by extracting key time control features from the time sequence distribution. The turning points of each adjustment period in the time sequence distribution are analyzed, and the turning points correspond to the time when the adjustment strategy switches. The turning point from the rapid adjustment period to the stable adjustment period is identified, which marks that the initial large disturbance is effectively suppressed and enters the steady state adjustment stage. The turning point from the stable adjustment period to the fine adjustment period is identified, which marks that the deviation has approached the target value and enters the accurate alignment stage. The control feature parameters at each turning point are extracted, including the deviation value, gain configuration, stability margin and response speed at that time. In addition to the turning points, other key times in the time sequence distribution are identified, including the maximum deviation time, the maximum gain time, the minimum stability margin time and the fastest response speed time. The control feature parameters of these key times are extracted to form a complete feature set together with the turning point features. Cluster analysis is performed on the feature set, and time points with similar features are classified into the same category. The clustering results show several typical control modes, such as rapid response mode, stable tracking mode and accurate maintenance mode. The mapping relationship between the control mode and the feature parameter is established, and the control feature map is drawn. The control feature map takes time as the horizontal axis, control feature parameters as the vertical axis, and different colors and markers are used to distinguish different control mode regions.

[0075] Step S150, interval mapping fusion of the attitude compensation constraint matrix and the compensation gain sequence is generated to generate a fusion control law, based on the fusion control law and the abnormal disturbance label to determine the dynamic priority control coefficient, and the dynamic priority control coefficient is combined with the control feature map to output the target continuous locking attitude control instruction.

[0076] Specifically, the attitude compensation constraint matrix and the compensation gain sequence are interval mapping fused to generate a fusion control law. The time axis of the compensation gain sequence is divided into multiple control intervals, and each interval corresponds to a relatively stable gain configuration. The basis for interval division is the change rate of the gain value. When the gain change rate of adjacent time is less than the threshold 0.2 per second, it is classified into the same interval, and when the change rate exceeds the threshold, a new interval is divided. The representative gain value of each interval is extracted, and the representative gain is the weighted average value of the gain value in the interval, and the weight is determined according to the stability margin of each time. The constraint parameters of the attitude compensation constraint matrix are matched and fused with the representative gain of each interval. The constraint matrix specifies the upper and lower limit range and the coupling relationship of the gain, and the representative gain provides the actual gain configuration of the interval. The control law expression of each interval is calculated, and the control law adopts a PID control structure, and the expression is Where u(t) is the control output, e(t) is the error signal, K_p, K_i, K_d are the proportional, integral and derivative gains respectively. The gain parameters are taken from the representative gain configuration of the interval, and are fine-tuned according to the attitude compensation constraint matrix to ensure that the constraint conditions are met. A smooth transition function is set at the boundary of adjacent intervals to avoid sudden changes in the control law parameters leading to jumps in the control output. The transition function adopts an S-shaped curve to realize smooth switching from the control law of the previous interval to the control law of the next interval within a transition period of 0.2 seconds. The control law expressions of each interval are combined in time sequence to form a complete fusion control law.

[0077] The dynamic priority control coefficient is determined based on the fusion control law and the abnormal disturbance label. The abnormal events recorded in the abnormal disturbance label are analyzed to identify the fusion control law interval corresponding to the abnormal occurrence time. In the abnormal occurrence interval, the control priority is adjusted according to the abnormal type in the abnormal disturbance label. For the over-response type of abnormality, the control priority coefficient in this interval is reduced to between 0.7 and 0.85 to weaken the response strength of the control output. For the under-response type of abnormality, the control priority coefficient in this interval is increased to between 1.15 and 1.3 to enhance the response strength of the control output. The control priority coefficient is a dimensionless value, and the reference value is set to 1.0. The specific value of the coefficient is dynamically determined according to the deviation amplitude in the abnormal disturbance label. The larger the deviation amplitude, the larger the adjustment range of the coefficient. For the normal interval without marked abnormalities, the control priority coefficient maintains the reference value of 1.0. At the junction of the normal interval and the abnormal interval, a gradual transition of the priority coefficient is set, and the transition length is set to 0.3 seconds. The comprehensive control output at each time is calculated, which is equal to the basic control output calculated by the fusion control law multiplied by the dynamic priority control coefficient. The dynamic priority control coefficient realizes dynamic adjustment of the control strength, enabling the control to adaptively adjust the response strategy according to historical abnormal information.

[0078] The target persistent lock attitude control command is output using the dynamic priority control coefficient combined with the control feature map. The control mode category corresponding to the current time is extracted from the control feature map, which indicates whether the current stage is fast response, smooth tracking or precise holding. According to the control mode category, the corresponding control strategy template is selected. The fast response mode adopts a large step size fast adjustment strategy, the smooth tracking mode adopts a moderate step size stable tracking strategy, and the precise holding mode adopts a small step size fine adjustment strategy. The dynamic priority control coefficient at the current time is read, and the output amplitude of the control strategy indicated by the control feature map is scaled by the coefficient. The three components of the attitude control command, pitch angle control command, yaw angle control command and height control command, are calculated. The pitch angle control command is calculated based on the pitch deviation, control mode in the control feature map and dynamic priority control coefficient, and the dimension of the command value is degree / second. Considering the coupling and coordination between multiple degrees of freedom, when pitch and yaw need to be adjusted at the same time, the coupling constraint coefficient of the attitude compensation constraint matrix is used to coordinate the control amplitudes of the two directions. The calculated control command is subjected to amplitude limiting processing, the upper limit of the pitch angle rate is ±5 degrees / second, the upper limit of the yaw angle rate is ±8 degrees / second, and the upper limit of the height change rate is ±2 meters / second. Rate smoothing filtering is performed, and a first-order low-pass filter is used to smooth the control command sequence. The filter time constant is set to 0.1 seconds to eliminate high-frequency jitter components in the command. The smoothed target persistent lock attitude control command is output, which includes the expected change rate of the pitch, yaw and height channels, and is sent to the flight control and gimbal control of the unmanned aerial vehicle to realize persistent and stable locking of the ground target.

[0079] To perform a kind of unmanned aerial vehicle to the posture control method of target persistent locking corresponding to above-mentioned method embodiment, to realize corresponding function and technical effect. Figure 2 , Figure 2 The structure block diagram of the posture control system 200 for the target persistent locking of unmanned aerial vehicle provided by the embodiment of the application is shown.The part related to the embodiment is shown only for the convenience of description, and the posture control system 200 for the target persistent locking of unmanned aerial vehicle provided by the embodiment of the application includes:

[0080] The data acquisition module 201 is configured to acquire position data stream of ground target and attitude data stream of unmanned aerial vehicle, perform coordinate conversion on the position data stream and the attitude data stream to generate target attitude relationship identifier, and perform deviation vector decomposition on the target attitude relationship identifier to generate visual axis deviation feature set.

[0081] The trajectory identification module 202 is configured to identify attitude correction trajectory by performing attitude mapping processing on the visual axis deviation feature set, extract control feature mode from the attitude correction trajectory to perform adaptive association to generate response feature factor, and construct attitude compensation constraint matrix based on the response feature factor.

[0082] The disturbance analysis module 203 is configured to determine coupling correlation of different disturbances by performing disturbance behavior analysis on the visual axis deviation feature set, identify superimposed disturbance sequence by using the coupling correlation, and perform dynamic reconstruction on the attitude compensation constraint matrix according to the superimposed disturbance sequence to generate compensation gain sequence.

[0083] The stability evaluation module 204 is configured to determine time sequence distribution of attitude adjustment by performing stability evaluation on the compensation gain sequence, identify abnormal disturbance feature from the time sequence distribution to generate abnormal disturbance label, and extract key time control feature from the time sequence distribution to obtain control feature spectrum.

[0084] The instruction generation module 205 is configured to perform interval mapping fusion on the attitude compensation constraint matrix and the compensation gain sequence to generate fusion control law, determine dynamic priority control coefficient based on the fusion control law and the abnormal disturbance label, and output target persistent locking posture control instruction by using the dynamic priority control coefficient in combination with the control feature spectrum.

[0085] The above-mentioned posture control system 200 for the target persistent locking of unmanned aerial vehicle can implement the posture control method for the target persistent locking of unmanned aerial vehicle of the above-mentioned method embodiment.The optional items in the above-mentioned method embodiment are also applicable to the embodiment, and will not be described in detail here.The remaining content of the embodiment of the application can be referred to the content of the above-mentioned method embodiment, and will not be described in detail in the embodiment.

[0086] The above embodiments are also not exhaustive enumeration based on the present application, in addition to which there can be a plurality of other embodiments not listed. Any substitution and improvement made without violating the concept of the present application is within the scope of the present application.

Claims

1. A method for controlling the attitude of a UAV for persistent lock-on to a ground target, characterized in that, The method comprises the following steps: acquiring a position data stream of a ground target and an attitude data stream of a UAV, performing coordinate conversion on the position data stream and the attitude data stream to generate a target attitude relationship identifier, and performing bias vector decomposition on the target attitude relationship identifier to generate a line-of-sight bias feature set; performing attitude mapping processing on the line-of-sight bias feature set to identify an attitude correction trajectory, including: decomposing the line-of-sight bias feature set into a fast response chain and a slow response chain; sending a trajectory connection marker from the fast response chain to the slow response chain; collecting the position information of the broken nodes of the trajectory connection marker to form a broken node list; selecting the path with the highest reconnection probability at the broken part as the attitude correction trajectory based on the broken node list; extracting a control feature mode from the attitude correction trajectory to perform adaptive association to generate a response feature factor, and constructing an attitude compensation constraint matrix based on the response feature factor; performing perturbation behavior analysis on the line-of-sight bias feature set to determine the coupling correlation degrees of different perturbations, including: performing frequency domain decomposition on the line-of-sight bias feature set to obtain a frequency band amplitude distribution spectrum; performing cross-coupling calculation on the frequency band amplitude distribution spectrum to generate a mutual coupling response channel; extracting a response transmission feature from the mutual coupling response channel as a perturbation coupling feature; summarizing the perturbation coupling features into coupling correlation degrees of different perturbations; using the coupling correlation degrees to identify a superimposed perturbation sequence, and dynamically reconstructing the attitude compensation constraint matrix according to the superimposed perturbation sequence to generate a compensation gain sequence, including: performing perturbation aggregation intensity analysis on the superimposed perturbation sequence to locate an intensity aggregation domain; generating a jump marker based on the intensity aggregation domain; performing gradient-driven reconstruction on the jump marker to obtain a potential gain candidate; determining a compensation gain sequence based on the potential gain candidate and the attitude compensation constraint matrix; performing stability evaluation on the compensation gain sequence to determine the time sequence distribution of attitude adjustment, identifying an abnormal perturbation feature from the time sequence distribution to generate an abnormal perturbation label, and extracting a key time control feature from the time sequence distribution to obtain a control feature map; performing interval mapping fusion on the attitude compensation constraint matrix and the compensation gain sequence to generate a fusion control law, determining a dynamic priority control coefficient based on the fusion control law and the abnormal perturbation label, and outputting a target continuous locking attitude control instruction by using the dynamic priority control coefficient in combination with the control feature map.

2. The method of claim 1, wherein, The bias vector decomposition on the target attitude relationship identifier to generate the line-of-sight bias feature set comprises the following steps: performing shaft system projection analysis on the target attitude relationship identifier to obtain a three-axis bias component set; performing amplitude scale calibration on the three-axis bias component set to determine a bias dominant axis direction; converting the bias dominant axis direction into an attitude adjustment reference point to form a reference point guide table; establishing a line-of-sight bias feature set according to the reference point guide table.

3. The method of claim 1, wherein, The extraction of the control feature mode from the attitude correction trajectory to perform adaptive association to generate a response feature factor comprises the following steps: extracting a control feature mode from the attitude correction trajectory to construct a response relationship graph; A matching degree measurement is performed on the response relationship graph to obtain a trajectory matching degree matrix; Weak coupling extraction is performed based on the trajectory matching degree matrix to obtain a weak coupling collaborative superposition factor; The cumulative intensity value of the weak coupling collaborative superposition factor is used as a response characteristic factor.

4. The method of claim 1, wherein, The abnormal disturbance feature is identified from the time sequence distribution to generate an abnormal disturbance label, including: The time sequence distribution is subjected to multi-layer mapping conversion to generate a mapping extreme value set; The abnormal disturbance feature is identified from the mapping extreme value set to generate a deviation sign; Deviation compensation processing is performed based on the deviation sign to generate a compensation disturbance index string; The compensation disturbance index string and the mapping extreme value set are fused and arranged as an abnormal disturbance label.

5. The method of claim 1, wherein, The trajectory connection marker is collected to form a fracture node list, including: The connection fracture critical point in the trajectory connection marker is analyzed; Fracture mode recognition is performed on the connection fracture critical point to generate a fracture feature set; The fracture period attribute is extracted with the help of the fracture feature set to obtain a period configuration matrix; The connection fracture critical point is subjected to fracture mark labeling based on the period configuration matrix to generate a fracture node list.

6. The method of claim 3, wherein, The weak coupling collaborative superposition factor is obtained by weak coupling extraction based on the trajectory matching degree matrix, including: Gradient distribution analysis is performed on the trajectory matching degree matrix to obtain a gradient jump distribution; Jump point recognition is performed on the gradient jump distribution to generate a jump point group; The jump amplitude attribute is extracted with the help of the jump point group to form an amplitude attribute vector; The weak coupling collaborative superposition factor is obtained by performing collaborative superposition amplification based on the amplitude attribute vector.

7. An attitude control system for persistent lock of unmanned aerial vehicles on ground targets, characterized by, It includes: A data acquisition module is configured to acquire a position data stream of a ground target and an attitude data stream of a UAV, perform coordinate conversion on the position data stream and the attitude data stream to generate a target attitude relationship identifier, and perform deviation vector decomposition on the target attitude relationship identifier to generate a line-of-sight deviation feature set; A trajectory identification module is configured to identify an attitude correction trajectory by performing attitude mapping processing on the line-of-sight deviation feature set, including: disassembling the line-of-sight deviation feature set into a fast response chain and a slow response chain; sending a trajectory connection marker from the fast response chain to the slow response chain; collecting fracture node position information of the trajectory connection marker to form a fracture node list; selecting a path with the highest reconnection probability at a fracture as an attitude correction trajectory based on the fracture node list; extracting a control feature mode from the attitude correction trajectory to perform adaptive association to generate a response characteristic factor, and constructing an attitude compensation constraint matrix based on the response characteristic factor; The disturbance analysis module is used for determining coupling degrees of different disturbances for the disturbance behavior analysis of the set of visual axis deviation features, including: performing frequency domain decomposition on the set of visual axis deviation features to obtain a frequency band amplitude distribution spectrum; performing cross-coupling calculation on the frequency band amplitude distribution spectrum to generate a mutual coupling response channel; extracting response transmission features on the mutual coupling response channel as disturbance coupling features; summarizing the disturbance coupling features as coupling degrees of different disturbances; using the coupling degrees to identify a superimposed disturbance sequence, and dynamically reconstructing the attitude compensation constraint matrix according to the superimposed disturbance sequence to generate a compensation gain sequence, including: performing disturbance aggregation intensity analysis on the superimposed disturbance sequence to locate an intensity aggregation region; generating a jump mark based on the intensity aggregation region; performing gradient-driven reconstruction on the jump mark to obtain a potential gain candidate; determining a compensation gain sequence based on the potential gain candidate and the attitude compensation constraint matrix; The stability evaluation module is used for performing stability evaluation on the compensation gain sequence to determine a time sequence distribution of attitude adjustment, identifying abnormal disturbance features from the time sequence distribution to generate an abnormal disturbance label, and extracting key time control features from the time sequence distribution to obtain a control feature map; The instruction generation module is used for performing interval mapping fusion on the attitude compensation constraint matrix and the compensation gain sequence to generate a fusion control law, determining a dynamic priority control coefficient based on the fusion control law and the abnormal disturbance label, and outputting a target continuous locking attitude control instruction by using the dynamic priority control coefficient in combination with the control feature map.

Images