Anti-interference control method applied to unmanned aerial vehicle hoisting operation in narrow scene

Abstract

The application relates to the technical field of unmanned aerial vehicle control, in particular to an anti-interference control method for unmanned aerial vehicle hoisting operation in a narrow scene, which comprises the following steps: collecting target object motion data and unmanned aerial vehicle poses in the flight process of the unmanned aerial vehicle hoisting the target object to a live wire, obtaining pose difference of the target object relative to the live wire based on the data, determining an interference detection time window from a pose difference change sequence at each detection time point; determining an optimal filter based on the interference detection time window to filter sensing data, and outputting flight control parameters by using the filtered data through an MPC algorithm; redetermining the interference detection time window at a next detection time point, and adaptively updating the time step number of the MPC algorithm according to the length growth amount of adjacent windows. The application can dynamically adapt to an electromagnetic interference environment, considers control stability and operation efficiency, and realizes stable anti-interference control of unmanned aerial vehicle hoisting operation in a narrow scene.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and more specifically to an anti-disturbance control method for UAV hoisting operations in confined spaces. Background Technology

[0002] In live overhead conductor operations, when it is necessary to freely guide the conductor to the break point, bridging is typically used to complete the electrical connection. Conventional bridging operations rely on insulated boom lifts as lifting platforms to transport personnel and bridging tools (such as bridging robots) to the vicinity of the conductor for connection. However, in confined areas with significant site constraints, such as under viaducts, narrow alleyways, fields, and hilltops, insulated boom lifts cannot enter or deploy, rendering bridging operations impossible.

[0003] To address this, existing technologies utilize drones instead of insulated boom trucks to hoist the bridging tools to the vicinity of the conductor and complete the connection. The drones, equipped with data sensing devices such as lidar and inertial measurement units (IMUs), and employing control algorithms like model predictive control (MPC) to generate flight control parameters, achieve automated hoisting.

[0004] However, there are strong electromagnetic fields around charged wires, and even corona discharge and spark discharge interference. These electromagnetic interferences can couple to the signal acquisition link of data sensing devices (such as A / D modules or timing modules), causing a large amount of noise to be mixed into the data acquired by the data sensing devices, which seriously affects the data quality and the efficient and stable control process of UAVs. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an anti-disturbance control method for drone hoisting operations in confined spaces.

[0006] The anti-disturbance control method for UAV hoisting operations in confined spaces, as described in this invention, adopts the following technical solution: One embodiment of the present invention provides an anti-disturbance control method for UAV hoisting operations in confined spaces, the method comprising the following steps: During the flight of the UAV carrying the target object toward the charged conductor, IMU sensors are used to collect the target object's motion data. Based on the UAV's pose and the target object's motion data, the flight parameters are controlled using the MPC algorithm, with the time step count in the MPC algorithm initialized to a preset value. The pose difference between the target object and the charged conductor is obtained based on the UAV's pose and the target object's motion data. At any given detection time point, an interference detection time window is determined from the sequence of pose difference changes to detect the amplitude of electromagnetic interference. Based on the interference detection time window, the optimal filter is determined for filtering the UAV's pose and the target object's motion data. After this detection time point, the UAV's pose and the target object's motion data are filtered using the optimal filter, and the flight parameters are controlled using the MPC algorithm. At the next detection time point after this detection time point, the interference detection time window is re-determined based on the sequence of pose difference changes, and the time step count of the MPC algorithm is updated using the increase in the window length of the two most recently obtained interference detection time windows.

[0007] Preferably, the specific steps included in updating the time steps of the MPC algorithm are as follows: The increase in window length of the two most recently obtained interference detection time windows is denoted as d1, and the absolute value of the mean of the motion trend amplitude of the two most recently obtained interference detection time windows is denoted as d2. When d1 is greater than or equal to 0, the updated time step is positively correlated with d1 and d2; when d1 is less than 0, the updated time step is set to the initial value of the time step.

[0008] Preferably, the specific steps for obtaining the interference detection time window are as follows: D1: The pose differences at all times within the historical time period before any detection time point constitute a change sequence, and any motion component in the change sequence is denoted as the target sequence; the target sequence is normalized and a value K0 is initialized at the same time; D2: The last K0 elements of the linearly normalized target sequence are denoted as the first group of elements. The mean and standard deviation of the first group of elements are denoted as U1 and A1, respectively. The K0 elements before the first group of elements in the target sequence are denoted as the second group of elements. The mean of the second group of elements is denoted as U2, and the mean of the K0 elements before the second group of elements in the target sequence is denoted as U3. (|U1-U2|+|U2-U3|) / 2 is denoted as the first change magnitude of the target sequence, and |U1-U3| is denoted as the second change magnitude. || represents the absolute value. D3: When both the first and second change amplitudes are greater than A1, the time span of K0 elements is used as an interference detection time window; when the first or second change amplitude is less than or equal to A1, the value of K0 is increased, and D2 and D3 are repeated until an interference detection time window is obtained and then the process stops.

[0009] Preferably, the method for obtaining the motion trend amplitude of the interference detection time window is as follows: During the execution of D3, once an interference detection time window is obtained, U1-U3 is recorded as the motion trend amplitude corresponding to the interference detection time window.

[0010] Preferably, the specific steps for determining the optimal filter when filtering the UAV pose and target object motion data based on the interference detection time window are as follows: Each component sequence of UAV pose or target object motion data within a historical time period is assigned a filter; the filtered pose difference of the target object relative to the charged wire is obtained by filtering each component sequence using each filter; All filter pose differences within the historical time period constitute the filter change sequence; any motion component in the filter change sequence is denoted as the target filter sequence, and the standard deviation of the elements within the interference detection time window in the target filter sequence is denoted as the electromagnetic interference amplitude after filtering; the mean of the electromagnetic interference amplitudes corresponding to all motion components in the filter change sequence is denoted as the comprehensive interference amplitude after filtering. The filter corresponding to each component sequence with the smallest overall interference amplitude after filtering is taken as the optimal filter.

[0011] Preferably, the filtering pose difference of the target object relative to the charged wire includes the following specific steps: All component sequences include: the position in the UAV pose within the historical time period constitutes the first position component sequence, the attitude angle in the UAV pose constitutes the first attitude angle component sequence, the linear acceleration in the target object motion data constitutes the first acceleration component sequence, and the attitude angle in the target object motion data constitutes the second attitude angle component sequence. By sequentially filtering each component sequence using the filters corresponding to all component sequences, we obtain each filtered pose of the UAV and each filtered motion data of the target object. Based on the filtered pose of the UAV and the filtered motion data of the target object, we determine the difference in filtered pose of the target object relative to the charged wire.

[0012] Preferably, the step of selecting the filter corresponding to each component sequence with the minimum overall interference amplitude after filtering as the optimal filter includes the following specific steps: Each component sequence corresponds to a low-pass filter. A set of candidate values ​​is preset for the cutoff frequency of each low-pass filter. Each low-pass filter selects one of the candidate values. The candidate values ​​selected by the low-pass filters corresponding to all component sequences constitute a combination result. All low-pass filters filter each component sequence based on each combination result to obtain a filtered comprehensive interference amplitude. Traversing all combination results, the combination result with the minimum filtered comprehensive interference amplitude is taken as the optimal parameter of the low-pass filters corresponding to all component sequences. All low-pass filters under the optimal parameter form the optimal filter.

[0013] Preferably, the steps of obtaining the pose difference of the target object relative to the charged conductor and determining the filtered pose difference of the target object relative to the charged conductor are both implemented in the following ways: Based on the point cloud on the charged conductor collected in real time by the UAV, a reference coordinate system is established with the charged conductor as the origin; the UAV pose or the filtered pose of the UAV includes: the position and attitude angle of the UAV in the reference coordinate system; the target object motion data or the filtered motion data of the target object includes: the linear acceleration and attitude angle of the target object. The position of the target object relative to the UAV is determined based on the linear acceleration and the cycloid length when the target object is hoisted on the UAV. The position of the target object relative to the UAV and the position of the UAV in the reference coordinate system are used to determine the position of the target object relative to the charged wire. The attitude angle of the target object relative to the charged wire is determined based on the attitude angle of the target object and the attitude angle of the UAV in the reference coordinate system. The position and attitude angle of the target object relative to the charged wire are denoted as the pose of the target object relative to the charged wire, and are used as the pose difference or the filtered pose difference.

[0014] Preferably, any one motion component in the changing sequence is denoted as the target sequence, and the specific steps include the following: Each element of the variation sequence contains two motion components: position and attitude angle. The position of each element is taken as the magnitude and recorded as the position difference of each element. The attitude angle of each element is taken as the magnitude and recorded as the attitude difference of each element. The position differences of all elements in the variation sequence constitute a position difference sequence, and the attitude differences of all elements constitute an attitude difference sequence. Any one of the position difference sequence and the attitude difference sequence is represented as the target sequence.

[0015] Preferably, the specific steps of controlling flight parameters based on the UAV pose and target object motion data, and using the MPC algorithm, are as follows: The pose of the target object relative to the charged conductor is determined based on the pose of the UAV and the motion data of the target object. The pose of the target object relative to the charged conductor is used as the state vector of the MPC algorithm, and the three-dimensional linear velocity and angular velocity of the UAV around the direction of gravity are used as flight parameters.

[0016] The beneficial effects of the technical solution of the present invention are: This invention dynamically assesses the amplitude of electromagnetic interference (EMI) through an interference detection time window and adaptively determines the optimal filter based on the interference detection results. This effectively suppresses sensor noise under the current EMI environment, ensuring the accuracy of the input MPC algorithm. Furthermore, by utilizing the increase in the window length of adjacent interference detection time windows, the time step count of the MPC algorithm is adaptively updated. When EMI is significant, the time step count is updated to a larger value, extending the prediction time domain to make the control strategy smoother, ensuring the stability of the UAV hoisting process, reducing the impact of non-EMI factors on EMI assessment, and improving filtering reliability. When EMI is weak, the time step count is updated to a smaller value, shortening the prediction time domain to improve control response speed and increase hoisting operation efficiency.

[0017] This invention forms a closed-loop anti-interference mechanism that integrates interference perception, filtering and suppression, and control strategy optimization. It can dynamically adapt to the changing electromagnetic interference environment near charged conductors, taking into account both control stability and operational efficiency, and achieving efficient and stable anti-interference control for UAV hoisting operations in confined spaces. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of an anti-disturbance control method for drone hoisting operations in confined spaces, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a hoisting device provided in one embodiment of the present invention. Detailed Implementation

[0019] The following description, in conjunction with the accompanying drawings, details the specific solution of the anti-disturbance control method for UAV hoisting operations in confined spaces provided by this invention.

[0020] Example 1: Please see Figure 1 The diagram illustrates a flowchart of the anti-disturbance control method for UAV hoisting operations in confined spaces, as provided in Embodiment 1 of the present invention. The method includes the following steps: Step S101: During the process of the UAV lifting the target object and flying towards the live wire, the motion data of the target object is collected by the IMU sensor. Based on the UAV pose and the motion data of the target object, the flight parameters are controlled by the MPC algorithm. The time step in the MPC algorithm is initialized to a preset value. The pose difference of the target object relative to the live wire is obtained based on the UAV pose and the motion data of the target object.

[0021] During the flight of a drone carrying a target object (such as a bridging robot) towards a live conductor, point cloud data is collected by a point cloud acquisition device to determine the drone's pose (including position and attitude angles) relative to the conductor. Simultaneously, an IMU sensor is installed on the target object to acquire its motion data (including linear acceleration and attitude angles). Combining these two data points, the pose difference (including position and attitude differences) of the target object relative to the live conductor is calculated.

[0022] This embodiment employs the MPC algorithm instead of the traditional PID algorithm to control flight parameters. This is because the target object being hoisted has a large mass, and sudden gusts of wind may occur in confined spaces, causing swaying during hoisting flight and resulting in simultaneous deviations of the UAV in multiple degrees of freedom. The MPC algorithm solves for the optimal flight parameters of the UAV by predicting the state data of the next Np steps (Np being the initial result of the MPC algorithm's time step count, with a value of 3). It can simultaneously coordinate linear and angular velocity control across multiple dimensions. Compared to the single-loop independent control of the PID algorithm, it is more suitable for handling control scenarios involving multivariable coupling, large inertia, and external disturbances during hoisting operations.

[0023] Step S102: Determine the interference detection time window for detecting the amplitude of electromagnetic interference from the sequence of changes in pose difference at any detection time point.

[0024] The fluctuation characteristics in the pose difference change sequence are a direct reflection of the strength of electromagnetic interference (EMI). However, these fluctuation characteristics simultaneously contain two components: one is the random fluctuation caused by EMI (EMI component), and the other is the motion change of the target object as it approaches the charged conductor (motion component). Therefore, it is impossible to measure the EMI based solely on the fluctuation characteristics of the entire sequence. Thus, this embodiment defines an interference detection time window on the pose difference change sequence to detect the amplitude of EMI. The changes within this time window are primarily contributed by EMI, which helps to initially separate the EMI component from the motion component in the time dimension. The aforementioned EMI amplitude refers to the local fluctuation characteristics of the change sequence within the interference detection time window (e.g., the local amplitude of pose difference).

[0025] The length of the interference detection time window reflects the significance of the electromagnetic interference-induced fluctuations relative to the approach trend of the target object. The longer the window, the more dominant the electromagnetic interference fluctuations are over a longer time span, and the approach trend of the target object is overwhelmed by the interference. The shorter the window, the more dominant the approach trend of the target object is, and the weaker the interference effect is.

[0026] Step S103: Determine the optimal filter for filtering UAV pose and target motion data based on the interference detection time window. After this detection time point, use the UAV pose and target motion data filtered by the optimal filter and the MPC algorithm to control the flight parameters.

[0027] Since the interference detection time window can determine the electromagnetic interference amplitude used to describe the strength of electromagnetic interference, this embodiment designs a corresponding optimal filter based on the interference detection time window to filter the UAV pose and target object motion data to suppress noise in the current electromagnetic interference environment. In subsequent time intervals, the determined optimal filter is used for further filtering to ensure that the subsequently filtered UAV pose and target object motion data no longer contain a large amount of electromagnetic interference noise, making the MPC algorithm control more accurate.

[0028] Step S104: At the next detection time point after the current detection time point, the interference detection time window is re-determined based on the change sequence of pose difference, and the time step number of the MPC algorithm is updated using the window length growth of the two most recently obtained interference detection time windows.

[0029] The above method determines an interference detection time window at a detection time point and determines the optimal filter based on it. At the next detection time point, the interference detection time window is redefined (and the optimal filter is redefined based on the interference detection time window).

[0030] Regarding the window length increases of the two recently obtained interference detection time windows, the increase in window length characterizes the change in the significance of electromagnetic interference before and after filtering. A larger increase in window length means that electromagnetic interference is actually more pronounced after filtering. This is because if electromagnetic interference is effectively suppressed after filtering, the interference component in the pose difference change sequence should decrease and the window should narrow. The fact that the window is longer instead indicates that the interference component has not been effectively filtered out, suggesting that the filter may have failed. The root cause is that non-interference factors such as the target object's running turbulence (residual electromagnetic interference noise in historical pose data causes MPC control instability, leading to running turbulence) or the target object's approach trend may have been mixed into the previous interference detection time window, resulting in inaccuracy of the interference detection time window used to detect the amplitude of electromagnetic interference. Conversely, a smaller increase in window length indicates that filtering effectively suppressed interference.

[0031] When the window length increases significantly (significant electromagnetic interference), even redetermining the optimal filter may not effectively suppress the interference because the root cause lies in the inaccurate acquisition of the interference detection time window. In this case, this embodiment further suppresses the impact of electromagnetic interference on the hoisting process by changing the UAV's control strategy. Specifically, the time step count of the MPC algorithm is set to a larger value, extending the prediction time domain and making the control strategy smoother and less aggressive, thus maintaining stable UAV movement even when sensor data still contains residual noise. Stable movement, in turn, makes the subsequent pose difference sequence cleaner, reducing the intrusion of non-electromagnetic interference factors, making the evaluation of the interference detection time window more accurate, and thus improving the reliability of the filter. When the window length increases only slightly (electromagnetic interference is not significant and interference can still be effectively suppressed), the time step count is also set to a smaller value, shortening the prediction time domain, improving the control response speed, allowing the UAV to approach the guide wire faster, and improving the efficiency of the hoisting operation.

[0032] In summary, this embodiment utilizes a closed-loop mechanism formed by interference detection time window, filter design, window length growth feedback, and MPC time step adaptive update. This mechanism enables the UAV hoisting control process to perceive changes in the electromagnetic interference environment in real time, dynamically suppress electromagnetic interference, and adaptively adjust the control strategy. When interference is severe, priority is given to ensuring control stability, while when interference is weak, priority is given to ensuring efficiency. Ultimately, this further enables anti-interference control for UAV hoisting operations near live wires in confined spaces.

[0033] As a preferred implementation, based on the UAV pose and target object motion data, and using the MPC algorithm to control flight parameters, the following methods are included: The UAV uses point cloud data collected by a point cloud acquisition device to extract point clouds on the guide line to be operated on. The PCA algorithm is used to obtain the principal component directions (the directions of the principal components with the largest eigenvalues) of these point clouds. A right-handed coordinate system is established with this principal component direction as the Y-axis and the direction perpendicular to the Y-axis in the vertical plane as the Z-axis (upward is the positive direction), denoted as the reference coordinate system. The origin of the reference coordinate system is set at the midpoint of all point clouds on the guide line. For the sensor coordinate system of the point cloud acquisition device (which is a known quantity), the translation vector T and rotation matrix R to be solved are used to perform translation and rotation transformations on the reference coordinate system to obtain the transformed coordinate system. The translation vector and rotation matrix when the transformed coordinate system coincides with the sensor coordinate system are solved (for example, using the Gauss-Newton method). The attitude angles (yaw, pitch, and roll) in this translation vector and rotation matrix represent the pose (position and attitude angles) of the point cloud acquisition device relative to the guide line, that is, the position and attitude of the point cloud acquisition device in the reference coordinate system.

[0034] This process utilizes well-known coordinate system transformation knowledge, which will not be described in detail in this embodiment. The point cloud acquisition device described in this embodiment is a binocular camera with a frequency of 12Hz and a resolution of 1920×1080. In other embodiments, the point cloud acquisition device may also be a lidar.

[0035] In this embodiment, a hoisting coordinate system parallel to the sensor coordinate system is established with the hoisting point of the target object as the origin. For the three-axis IMU mounted on the target object (the IMU is considered the target object in this embodiment), its output is the linear acceleration and angular velocity of the target object, with a sampling frequency of 1kHz. The attitude angles (yaw, pitch, and roll) of the target object in the hoisting coordinate system are obtained by integrating the angular velocity, which represents the attitude of the target object. In other embodiments, the accuracy of the attitude angles can be further improved by performing Kalman filtering on the integral result of the angular velocity. It should be noted that the errors of all sensors are also treated as part of the electromagnetic interference and suppressed in subsequent processes.

[0036] Measure the cycloidal length L from the IMU to the sling point; determine the target object's position in the sling coordinate system (i.e., its position relative to the UAV) based on the target object's attitude angle, linear acceleration, and cycloidal length L; add the displacement from the sensor coordinate system to the sling coordinate system (obtained through measurement) to the target object's position in the sling coordinate system, and add the position of the point cloud acquisition device in the reference coordinate system to obtain the target object's position relative to the guide wire; add the target object's attitude angle to the point cloud acquisition device's pose angle relative to the guide wire to obtain the target object's pose angle relative to the guide wire; use the target object's position and attitude angle relative to the guide wire as the target object's pose relative to the guide wire.

[0037] Thus, the pose of the target object relative to the guide wire was obtained by real-time synchronous acquisition of the UAV pose and the target object pose. This process is based on conventional coordinate system transformation and geometric knowledge, and the principle of each step of the above process will not be elaborated in this embodiment.

[0038] Furthermore, based on the pose of the target object relative to the guide wire, the flight parameters are controlled using the MPC algorithm.

[0039] It should be noted that the IMU on the target object is installed under laboratory conditions. During installation, the three axes of the IMU are aligned with the three axes of the sensor coordinate system (the hoisting coordinate system) (keeping them parallel). In the laboratory (without wind, electromagnetic interference, etc.), the drone lifts the target object and hovers it. When the target object does not swing, the angular velocity and linear acceleration outputs of the three axes of the IMU are calibrated to 0.

[0040] The method for determining the position of the target object in the hoisting coordinate system based on the target object's attitude angle, linear acceleration, and cycloidal length L includes: Subtracting the acceleration of the drone from the linear acceleration of the target object yields the acceleration G0 of the target object relative to the drone. Since the motion of the target object relative to the drone (i.e., the motion in the hoisting coordinate system) is a simple pendulum motion with a pendulum length of L, a pendulum acceleration of G0, and a gravitational acceleration of g.

[0041] The method for obtaining the acceleration of the drone at each moment is as follows: the velocity at each moment is obtained based on the position at each moment and the previous moment, and the acceleration at each moment is obtained based on the velocity at each moment and the velocity at the previous moment. If there are less than two moments before each moment (i.e., the hoisting has just begun), the drone acceleration is not obtained, and the linear acceleration of the target object is directly taken as G0, which does not affect the subsequent hoisting process.

[0042] In the hoisting coordinate system, obtain the vertical plane containing the acceleration G0 (a three-dimensional vector), which is the pendulum plane. The magnitude of the oscillation acceleration within this plane is the modulus of G0, denoted as g0. The oscillation angle 'a' within the plane is the arcsine of g0 / g. Let 'b' denote the counterclockwise rotation angle between the pendulum plane and the XZ plane. Then, the coordinates of the IMU in the hoisting coordinate system are: .

[0043] All steps involved in this process are based on well-known physical and geometric principles, and will not be described in detail in this embodiment. It should be noted that when g0 is greater than or equal to g, it indicates that the motion of the target object is seriously abnormal (it is no longer a simple pendulum motion, and the target object may be subject to strong interference from strong winds). At this time, the drone should immediately return to base and choose a day with less wind for hoisting.

[0044] As an example, the methods for extracting point clouds from the traverse lines to be processed include: A marker, such as a red ribbon, is installed on the guide wire to be operated. The drone's binocular camera is pointed towards the guide wire to collect a set of point clouds. The RANSAC algorithm is used to fit all the straight lines in the point cloud set to obtain the interior points on each straight line. At the same time, the marker region is segmented from the image collected by the binocular camera. For the point cloud in the marker region, the distance from each point cloud to each straight line is calculated. The average distance of all point clouds in the marker region to each straight line is recorded as the first distance of each straight line. The straight line with the smallest first distance is obtained. This straight line represents the guide wire to be operated, and the interior points on this straight line are taken as all the point clouds on the guide wire to be operated.

[0045] In this embodiment, the distance threshold of the RANSAC algorithm is set to 1cm, and the maximum number of iterations is set to 300.

[0046] The method for segmenting the marker region is as follows: The R channel of the RGB image (e.g., the left eye image) captured by the binocular camera is obtained, and the region with a grayscale value greater than the segmentation threshold in the R channel is recorded as the marker region. In this embodiment, the grayscale value of red in the R channel is generally greater than 150. Therefore, the segmentation threshold is set to 150-200, preferably 180. Other values ​​can be used in other embodiments, and this embodiment does not impose specific limitations.

[0047] Preferably, based on the pose of the target object relative to the guide wire, the flight parameters are controlled using the MPC algorithm, including the following methods: After obtaining the pose (position and attitude angle) of the target object relative to the guide wire, the pose is dimensionless and concatenated into a six-dimensional vector, which is used as the state vector of the MPC algorithm. In this embodiment, the flight parameters include the three-dimensional linear velocity and angular velocity around the gravity direction of the UAV. These flight parameters are dimensionless and concatenated into a four-dimensional vector, which is used as the control vector of the MPC algorithm. The time step in the MPC algorithm is initialized to 3, that is, the MPC algorithm solves for the optimal flight parameters (that is, solves for the current optimal control vector) by predicting the state vectors of the next 3 steps.

[0048] Specifically, the state-control formula of the MPC algorithm is x(k+1)=Ax(k)+Bu(k), where x(k) represents the state vector at time k, u(k) represents the flight parameters at time k, x(k+1) represents the state vector at time k+1, and A and B represent the preset state matrix and control matrix.

[0049] Cost function of MPC algorithm .in , , As a preset weight, this embodiment needs to consider each dimension of both the state vector and the flight parameters simultaneously, therefore... , , All three are set to identity matrices, and are guaranteed to be identical.

[0050] Since the control objective of this embodiment is to align the target object with the wire, the target value of the state vector in the third step is... It is [0, 0, 0, 0, 0, 0] (which represents the position and attitude of the conductor in the reference coordinate system, where the position is the origin of the coordinate system and all attitude angles are 0). The MPC algorithm uses a convex optimization algorithm to minimize the cost function G, and then solves for the flight parameter u(k) at the k-th time, and uses u(k) to control the flight of the UAV.

[0051] The constraints required to minimize the cost function G in this embodiment are: the linear velocity of each dimension is -10 to 10 m / s, and the angular velocity around the direction of gravity is -3.5 to 3.5 rad / s.

[0052] It should be noted that in this embodiment, every 0.2 seconds is considered a moment, that is, every 0.2 seconds (i.e., at each moment) a state vector is obtained based on the most recently collected UAV pose and target object motion data, and then the MPC algorithm is used to adjust the UAV's flight parameters once.

[0053] As a preferred example, the method for removing dimensions from pose and flight parameters is as follows: Divide each dimension of position in the pose by 1m, divide each dimension of attitude angle in the pose by π, divide each dimension of linear velocity in the flight parameters by 1m / s, and divide the angular velocity in the flight parameters by 1rad / s. The results of these divisions are used as the dimensionless results.

[0054] Since removing dimensions is only for the convenience of the MPC algorithm, the control process of the MPC algorithm is not affected by any dimension removal method (that is, regardless of the value of the dividend). In other words, the implementation process and the implementation effect of this embodiment are not affected by the dimension removal method. Therefore, other embodiments can be divided by other values, and this embodiment does not impose specific limitations.

[0055] It should be noted that after the MPC algorithm solves for the optimal flight parameters, the dimensions of the optimal flight parameters need to be restored, that is, multiplied by 1 m / s or 1 rad / s respectively, before they can be used to control the UAV.

[0056] As an example, the state matrix A and control matrix B in the MPC algorithm are set up as follows: Before hoisting, conduct a manual flight test on the UAV (e.g., simulate the hoisting process in a laboratory or under applied working conditions) for one minute, recording the state vector and flight parameters at each moment. Use the state vector and flight parameters at each moment, as well as the state vector at the next moment, as a sample. All samples from the flight test constitute a sample set. Use the least squares method to fit the state-control formula of the MPC algorithm to obtain the parameters: state matrix A and control matrix B.

[0057] In all embodiments, the state matrix A and control matrix B only need to be fitted once, and once fitted, all embodiments can be used.

[0058] The above descriptions of the MPC algorithm are all well-known or conventional techniques, and will not be described in detail in this embodiment.

[0059] As a preferred implementation, the method for obtaining the pose difference of the target object relative to the charged wire based on the UAV pose and target object motion data includes: The pose difference is equal to the pose of the target object relative to the conductor minus the pose of the conductor. Since the pose of the conductor is [0, 0, 0, 0, 0, 0], the pose difference is equal to the pose of the target object relative to the conductor, which is a six-dimensional vector composed of position (three-dimensional vector) and attitude angle (three-dimensional vector).

[0060] As a preferred implementation, the interference detection time window for detecting the electromagnetic interference amplitude is determined from the sequence of pose difference changes at any detection time point, including the following methods: (1) The start time of this embodiment is when the UAV is in automatic control and the target object is fully lifted (that is, when the pressure between the target object and the transport vehicle is measured to be 0). The time from the start time to the current time is recorded as the historical time period.

[0061] In some embodiments, to reduce subsequent computation, the most recent 0.5 minutes can be used as the historical time period, and at the very least, the most recent 10 seconds can be used as the historical time period. If it is less than 10 seconds, it means that the drone has just started moving and no special control is required for the drone. At this time, the process and all steps in steps S102 to S104 are stopped (only step S101 is executed) until more than 10 seconds have passed before step S102 is started again.

[0062] The temporal sequence comprising the pose differences at all moments within a historical time period is denoted as the variation sequence. Each element of the variation sequence contains two motion components: position and attitude angle. The position difference of each element is calculated by taking the magnitude of its value, and the attitude difference of each element is calculated by taking the magnitude of its value. The position differences of all elements constitute a position difference sequence, and the attitude differences of all elements constitute an attitude difference sequence. Any sequence between the position difference sequence and the attitude difference sequence is denoted as the target sequence.

[0063] The target sequence is normalized (in this embodiment, linear normalization is used) to remove dimensions and orders of magnitude, and a value K0 is initialized at the same time.

[0064] (2) The last K0 elements of the linearly normalized target sequence are denoted as the first group of elements. The mean and standard deviation of the first group of elements are denoted as U1 and A1, the K0 elements before the first group of elements in the target sequence are denoted as the second group of elements, the mean of the second group of elements is denoted as U2, the K0 elements before the second group of elements in the target sequence are denoted as the third group of elements, and the mean of the third group of elements is denoted as U3.

[0065] In this embodiment, a detection time point is set every 10 to 30 moments. If the interval is too small (e.g., less than 10 moments), the amount of calculation will be too large and the power consumption will be too high. If the interval is too large (e.g., more than 30 moments), the interference detection time window for detecting the amplitude of electromagnetic interference cannot be obtained in time. In this embodiment, setting a detection time point every 15 moments is the optimal implementation method.

[0066] Let (|U1-U2|+|U2-U3|) / 2 be denoted as the first variation amplitude of the target sequence, and let |U1-U3| be denoted as the second variation amplitude. The first and second variation amplitudes represent the changes in position or attitude at two different time scales. In this embodiment, U1-U3 is denoted as the variation amplitude to describe the motion changes of the target object over a time span of K0 elements, and || represents the absolute value.

[0067] (3) When both the first and second variation amplitudes are greater than A1, the time span of K0 elements is taken as an interference detection time window. The fluctuation changes within this interference detection time window will not contain too much of the changing trend of position or attitude differences of the target object during hoisting, which helps to reliably assess the electromagnetic interference situation within the interference detection time window.

[0068] When the first or second change amplitude is less than or equal to A1, the time span of K0 elements is insufficient to analyze and evaluate electromagnetic interference, failing to consider more fluctuations in position or attitude differences, which may lead to inaccurate evaluation of electromagnetic interference amplitude. In this case, instead of using the time span of K0 elements as an interference detection time window, the value of K0 is increased, and processes (2) and (3) are repeated until an interference detection time window is obtained. In addition, if K0 reaches the preset upper limit value, the repeated execution is also stopped, and the current time span of K0 elements is used as an interference detection time window.

[0069] The time span is continuously increased during the repetition process to ensure that the fluctuation changes within the obtained interference detection time window are exactly smaller than the first and second change amplitudes. This ensures that the interference detection time window can take into account more fluctuations due to positional or orientation differences, and is as unaffected as possible by the movement trends of positional or orientation differences, thereby making the electromagnetic interference assessed within the interference detection time window as accurate as possible.

[0070] Once an interference detection time window is obtained, the corresponding change amplitude is recorded as the motion trend amplitude corresponding to the interference detection time window. This represents the motion amplitude of the target object moving toward the conductor within a local time span, with the interference detection time window as the time span. The motion trend of the target object toward the conductor within this time span is not overwhelmed or interfered with by electromagnetic interference.

[0071] Specifically, if there are fewer than K0 elements in the target sequence, or before the first group of elements, or before the second group of elements, before any interference detection time window has been acquired, it indicates that the UAV has just started moving and is in the early stage of hoisting. No special control of the UAV is required in the early stage. At this time, the process and all steps S102 to S104 are stopped. The process will resume from step S102 when the next detection time point arrives. If one or more interference detection time windows have been acquired, and the situation of fewer than K0 elements is encountered during the repeated execution of (2) and (3), the repeated execution is also stopped, and the time span of the current K0 elements is used as an interference detection time window.

[0072] The initial value of K0 is set to 5~13. In this embodiment, it is considered that if the initial value of K0 is too small, it will not have statistical characteristics, and if it is too large, it may take too much motion component into account. Therefore, in a preferred example, the initial value of K0 is set to 9. In addition, when increasing the value of K0 in this embodiment, the value of K0 is increased in step size of 2. This step size can increase the time span and avoid introducing motion component into the electromagnetic interference assessment due to the excessive increase of the time span each time.

[0073] To reduce computational load and avoid the aforementioned repetitive execution leading to an infinite loop, this embodiment sets an upper limit value for K0, preferably 20 to 40. This embodiment uses 30 as an example. When K0 is less than or equal to 30, it will not consume too much computation and can fully highlight the electromagnetic interference situation. When K0 is greater than 30, the time span of 30 elements will be directly used as an interference detection time window.

[0074] In summary, the above process obtains an interference detection time window for each of the two motion components: position difference and pose difference. Specifically, it obtains one interference detection time window based on the position difference sequence within the pose difference change sequence, and another based on the pose difference sequence within the pose difference change sequence. When the lengths of the interference detection time windows corresponding to these two motion components are different, only the longest interference detection time window is retained. This avoids quantization discrepancies in the interference detection time windows when updating the time steps later, ensuring that the MPC algorithm simultaneously considers the states of all motion components.

[0075] In this embodiment, the standard deviation of the elements in the interference detection time window of the target sequence will be denoted as the electromagnetic interference amplitude.

[0076] As a preferred implementation, the optimal filter for filtering UAV pose and target motion data is determined based on the interference detection time window, including the following methods: First, it should be noted that the drone pose includes the drone's position (a three-dimensional vector) and attitude angle (also a three-dimensional vector), while the target object motion data includes the IMU's linear acceleration (a three-dimensional vector) and attitude angle (also a three-dimensional vector).

[0077] The temporal sequence of positions formed by the UAV pose data within the historical time period is denoted as the first position component sequence, and the sequence of attitude angles formed by the UAV pose data is denoted as the first attitude angle component sequence. The temporal sequence of linear accelerations formed by the target object motion data within the historical time period is denoted as the first acceleration component sequence, and the sequence of attitude angles formed by the target object motion data is denoted as the second attitude angle component sequence.

[0078] In this embodiment, the four component sequences correspond to four numbered filters: first filter G1, second filter G2, third filter G3, and fourth filter G4. The first position component sequence is filtered using G1 to obtain filter sequence L1, the first attitude angle component sequence is filtered using G2 to obtain filter sequence L2, the first acceleration component sequence is filtered using G3 to obtain filter sequence L3, and the second attitude angle component sequence is filtered using G4 to obtain filter sequence L4.

[0079] Each element in L1 and each element in L2 constitutes each filtered pose of the UAV, and each element in L3 and each element in L4 constitutes each filtered motion data of the target object. Based on the filtered pose of the UAV and the filtered motion data of the target object, the filtered pose difference of the target object relative to the charged wire is determined (the process of obtaining the pose difference is the same as described above).

[0080] The time series sequence comprising all filtered pose differences within a historical time period is denoted as the filtered change sequence. Similarly, the filtered change sequence is also decomposed into two motion components: a position difference sequence and an attitude difference sequence (the specific method has been described above). Taking any one of these two sequences (denoted as the target filtered sequence) as an example, the following analysis is performed: The standard deviation of the elements in the interference detection time window of the target filtering sequence is denoted as the electromagnetic interference amplitude after filtering. It describes the electromagnetic interference situation remaining in the interference detection time window after the noise interference is suppressed by the filter (i.e., G1~G4).

[0081] The average value of the filtered electromagnetic interference amplitude corresponding to the two motion components of the position difference sequence and the attitude difference sequence is denoted as the filtered comprehensive interference amplitude. It reflects the electromagnetic interference situation remaining in the interference detection time window after comprehensively considering noise interference suppression under two different motion dimensions of displacement and attitude angle.

[0082] The above process, after assigning a set of filtering parameters (including the four filtering parameters of G1 to G4) to the four numbered filters, can realize the filtering process of UAV pose and target object motion data within a historical time period, and obtain a filtered comprehensive interference amplitude.

[0083] The set of filtering parameters that minimizes the overall interference amplitude after filtering is recorded as the optimal set of filtering parameters. Filters G1 to G4 under this set of optimal filtering parameters are used together as the optimal filter.

[0084] As an example, the set of filtering parameters that minimizes the overall interference amplitude after filtering is denoted as the optimal set of filtering parameters. The methods include: In this embodiment, the four numbered G1~G4 are all low-pass filters. For any one of the low-pass filters G1~G4, its filtering parameter is a cutoff frequency. A set of candidate values ​​is preset for the filtering parameter (i.e., the cutoff frequency) of the low-pass filter, and the corresponding component sequence (i.e., the filtering object) of the low-pass filter is recorded as the filtering sequence. The process of filtering using one of the filtering parameters (i.e., the cutoff frequency) is as follows: The amplitude spectrum and phase spectrum of the sequence to be filtered are obtained by using the Fourier transform algorithm. The amplitude values ​​greater than the cutoff frequency are set to 0 in the amplitude spectrum to obtain the cutoff amplitude spectrum. Based on the cutoff amplitude spectrum and phase spectrum, the inverse Fourier transform algorithm is used to transform to the time space to obtain the filtered sequence, which is any one of the above filtered sequences L1 to L4.

[0085] It is worth noting that each element in the sequence to be filtered is also a vector, that is, the sequence to be filtered is multi-channel (each dimension of the vector is a channel). The amplitude spectrum and phase spectrum mentioned above refer to the amplitude spectrum and phase spectrum of each channel, respectively. The filtering mentioned above refers to filtering the amplitude spectrum of all channels separately and then performing an inverse Fourier transform.

[0086] In addition, before performing a Fourier transform on the sequence to be filtered, multiple elements are interpolated using a linear interpolation algorithm before the first element and after the last element in each channel of the sequence to be filtered, respectively. In this embodiment, the number of interpolated elements is equal to half the length of the sequence to be filtered (rounded up). Then, the Fourier transform is performed. Note that after the inverse Fourier transform algorithm transforms to the time space, the interpolated elements are deleted to obtain the filtered sequence; this process is a conventional practice to avoid amplitude spectrum leakage.

[0087] For the four numbered low-pass filters G1~G4, the cutoff frequency of each low-pass filter corresponds to a set of candidate values. Each low-pass filter selects one of the candidate values, and the candidate values ​​selected by all low-pass filters constitute a combination result. Using this combination result, filtering is performed according to the above method to obtain the filter sequence L1~L4, and further, the comprehensive interference amplitude after filtering is obtained.

[0088] That is, each combination result corresponds to a filtered comprehensive interference amplitude. By iterating through all combination results, the combination result with the smallest filtered comprehensive interference amplitude among all combination results is obtained. This combination result is a set of optimal filtering parameters for G1~G4.

[0089] As an example, a set of candidate values ​​can be preset for the filtering parameters (i.e., cutoff frequency) of this low-pass filter, including the following methods: The optimal filter obtained at the previous detection time point is obtained. This optimal filter is also composed of the four numbered low-pass filters mentioned above. The cutoff frequency with the same number as the low-pass filter is obtained from these four filters and used as the reference frequency of the low-pass filter. A candidate interval is constructed with the reference frequency as the center. Several frequencies are sampled at equal intervals within the candidate interval as a set of candidate values.

[0090] Since the speed of the UAV is limited, its movement distance between adjacent detection time points will not change abruptly. Therefore, the electromagnetic interference between adjacent detection time points is similar. That is, the optimal cutoff frequency at the current detection time point is near the reference frequency, which is within the candidate interval. Therefore, this embodiment selects candidate values ​​from the candidate interval, which can greatly reduce the time to obtain the optimal filtering parameters.

[0091] As an example, the candidate interval can be set to [80%H, 120%H], where H represents the reference frequency. Within this interval, five frequencies are sampled at equal intervals as a group of candidate values. This method balances accuracy and computational efficiency. In other embodiments, the candidate interval and number of samples can be set to different values. For example, in embodiments with high computational power, the candidate interval length and number of samples can be increased to further ensure the accuracy of the optimal filtering parameters. Conversely, in embodiments with limited computational power, the candidate interval length and number of samples can be decreased to further improve computational efficiency, but this will reduce the accuracy of the optimal filtering parameters. Since the settings for the candidate interval and number of samples depend on the hardware's computational power, this embodiment does not impose specific limitations on their settings; it is sufficient to ensure that the number of candidate values ​​sampled within the candidate interval is greater than or equal to two.

[0092] Specifically, if there is no previous detection time point (at which point the drone is in the initial stage of hoisting), then for any one of the first position component sequence, first attitude angle component sequence, first acceleration component sequence, and second attitude angle component sequence within the historical time period, the STL algorithm is used to obtain the trend sequence of that sequence. The difference sequence between this sequence and the trend sequence contains high-frequency interference caused by electromagnetic signals. The Fourier transform algorithm is used to obtain the amplitude spectrum of the difference sequence, and the frequency with the largest amplitude in the amplitude spectrum is used as the reference frequency. Although this reference frequency may not be the optimal cutoff frequency, since it is in the initial stage of drone hoisting, using the largest frequency as the reference frequency has a smaller impact on the drone hoisting process. As the hoisting process progresses, it will gradually approach or equal the optimal cutoff frequency (that is, the optimal filtering parameter that can effectively suppress electromagnetic interference).

[0093] As a preferred implementation, the method following this detection time point, based on the UAV pose and target object motion data filtered by the optimal filter, includes: After obtaining the UAV pose and target motion data at any time after the detection time point, the corresponding first position component sequence, first attitude angle component sequence, first acceleration component sequence, and second attitude angle component sequence within the historical time period are obtained. These four component sequences are then filtered using the four numbered low-pass filters in the above-mentioned optimal filter.

[0094] Note that in the filtered sequences L1 to L4 obtained by filtering, only the last element is retained as the filtered UAV pose and target motion data obtained at any time after the detection time point.

[0095] As a preferred implementation, the time step count of the MPC algorithm is updated using the window length growth of the two most recently obtained interference detection time windows. This includes the following methods: In the above process, for two adjacent detection time points, an interference detection time window and the corresponding motion trend amplitude are determined based on the previous detection time point, which are respectively denoted as the first window and its motion trend amplitude. At the next detection time point, another interference detection time window and the corresponding motion trend amplitude are determined, which are respectively denoted as the second window and its motion trend amplitude.

[0096] Specifically, if there is no previous detection time point, then the subsequent process will not be executed, and the time step count of the MPC algorithm will not be updated.

[0097] It should be noted that in this embodiment, an optimal filter for subsequent filtering is determined based on the first window. At the next detection time point, after the second window is determined, the optimal filter for subsequent filtering is also re-determined based on it, so as to ensure the timely update of the optimal filter.

[0098] As described above in this embodiment, when the window length of the second window is greater than the window length of the first window, it indicates that even with the optimal filter, the electromagnetic interference component has not been effectively filtered out. The root cause is that the interference detection time window (i.e., the first window) used to determine the amplitude of electromagnetic interference may have been mixed with non-electromagnetic interference factors (i.e., motion factors) such as the target object's running turbulence (the electromagnetic interference noise remaining in the historical pose data causes MPC control instability, which in turn leads to running turbulence) or the target object's approach trend. This leads to the inaccuracy of the interference detection time window used to detect the amplitude of electromagnetic interference, which in turn causes the previous optimal filter to quickly fail. At this time, even if the optimal filter is re-determined based on the second window at the next detection time point, it may not be able to effectively suppress electromagnetic interference.

[0099] Based on this, this embodiment effectively suppresses electromagnetic interference by changing the control strategy of the UAV (i.e., the number of time steps to update the MPC algorithm).

[0100] Specifically, the difference between the window length of the second window and the window length of the first window is recorded as the window length growth, denoted as d1, and the absolute value of the mean of the motion trend amplitudes corresponding to the second window and the first window is recorded as d2.

[0101] When d1 is greater than or equal to 0, the updated time step number is positively correlated with d1 and d2. Here, d2 represents the average motion trend of the UAV during motion control at different time scales, which is not overwhelmed by electromagnetic interference (i.e., the average motion amplitude when approaching the guide wire).

[0102] The larger d1 is, the more the time step of the MPC algorithm needs to be set to a larger value in this embodiment. This extends the prediction time domain, resulting in a smoother and less aggressive control strategy, thus maintaining stable drone movement even when sensor data still contains residual noise. Smooth movement, in turn, makes the subsequent pose difference sequence cleaner, reduces the intrusion of non-interference factors, and makes the assessment of electromagnetic interference amplitude more accurate, thereby improving the reliability of filtering. Conversely, the smaller d1 is, the smaller the time step can be set, shortening the prediction time domain, improving control response speed, allowing the drone to approach the guide wire faster, and increasing operational efficiency.

[0103] Furthermore, a larger d2 indicates that the target object is approaching the conductor with a significant amplitude of movement within a local time, but the electromagnetic interference is significant and difficult to suppress (i.e., d1 is greater than or equal to 0). In this case, it is necessary to increase the number of time steps more to avoid large deviations in the trajectory of the drone or target object when it rapidly approaches the conductor due to the difficulty in estimating and suppressing electromagnetic interference (e.g., the drone rapidly moving to other positions due to electromagnetic interference), thus leading to low hoisting efficiency. When d1 is small, the amplitude of movement of the drone or target object is not significant within a local time (e.g., when approaching the conductor). In this case, the root cause of the small d1 may not be entirely due to electromagnetic interference. Although further increasing the number of time steps can still further suppress electromagnetic interference, it will reduce control efficiency (i.e., reduce hoisting efficiency). Therefore, this embodiment does not set the number of time steps too large to avoid low hoisting efficiency.

[0104] When d1 is less than 0, it means that the electromagnetic interference is not obvious or that the electromagnetic interference can continue to be suppressed. At this time, the time step is set to a smaller value, such as resetting to the initial value, so that the drone can approach the wire faster and improve the efficiency of the hoisting operation.

[0105] As an example, the formula for calculating the number of time steps to update the MPC algorithm is: When d1 is greater than or equal to 0 When d1 is less than 0, .

[0106] in, The initial value for the time step (equal to 3), d0 represents the window length of the first window, and the coefficient... Used to remove dimensions, where This represents the growth rate of the window length; This represents the updated time step (the final value is rounded to the nearest integer).

[0107] N0 represents the maximum increment of the time steps. In this embodiment, the maximum value of the time steps is 9. Exceeding this value results in a heavy processor load when running the MPC algorithm, leading to insufficient computing power and poor accuracy. Furthermore, this embodiment considers that when... When the value is greater than 1 (meaning the increase in window length is already greater than the length of the first window), it indicates that the increase in window length is relatively large. At this point, there is no need to consider further increases in window length. Therefore, when... When it is greater than 1, directly set That's fine. And because... It is also greater than or equal to 0 and less than or equal to 1. Therefore, in summary, N0=6 is the optimal implementation, which can ensure that the maximum number of time steps does not exceed 9, so as to adapt to the processor's computing power burden and control accuracy.

[0108] It should be noted that in some embodiments, the length of the interference detection time window has a maximum value (i.e., when K0=30 as mentioned above). When d0 is equal to or close to this maximum value (e.g., greater than 90% of the maximum value), the maximum value truncation operation causes d1 to be unable to accurately describe the change in the interference detection time window (at this time, the significance of electromagnetic interference cannot be reliably described). In this case, if d1 is greater than or equal to 0, directly set... Here, control efficiency is significantly sacrificed to suppress electromagnetic interference that may exist but cannot be further described and assessed.

[0109] In other embodiments, N0 can be set to other values ​​or other formulas can be used to update the time steps, depending on the specific circumstances. When other embodiments set N0 to other values ​​or use different time step update formulas, as long as the updated time steps are positively correlated with d1 and d2, the same technical effect as this embodiment can be achieved: when the interference is severe (d1 and d2 are large), the stability of the control is prioritized (the updated time steps are large); when the interference is weak (d1 and d2 are small), efficiency is prioritized (the updated time steps are small), and ultimately, anti-interference control of UAV hoisting operations near live wires in narrow scenarios is further realized. Therefore, this embodiment does not limit the value of N0 or the update formula used.

[0110] Example 2: In this embodiment, the IMU is installed in the following location: Figure 2 As shown, the bracket for mounting the IMU and the hoisting robot below, as a whole, constitute the target object described in the above embodiment.

[0111] The above embodiments simplify the target object as a point mass with position and orientation at the IMU, which can ensure that the position and orientation of the point mass are aligned with the live conductor. However, the purpose of this embodiment is to ensure that the hook below the IMU is hooked on the conductor. Therefore, when the above embodiments detect that the target object is close to the conductor, for example, when the modulus of the target object's position relative to the conductor is less than 3cm, a prompt message is sent to the construction personnel to indicate that it has approached the conductor. After receiving the message, the construction personnel switch to manual control mode. The construction personnel control the drone through the camera below the drone (a downward-facing camera) to ensure that the hook below the IMU is hooked on the conductor, thus ensuring the successful completion of the hoisting.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A disturbance rejection control method for UAV hoisting operations in confined spaces, characterized in that, The method includes the following steps: During the flight of the UAV carrying the target object toward the charged conductor, IMU sensors are used to collect the target object's motion data. Based on the UAV's pose and the target object's motion data, the flight parameters are controlled using the MPC algorithm, with the time step count in the MPC algorithm initialized to a preset value. The pose difference between the target object and the charged conductor is obtained based on the UAV's pose and the target object's motion data. At any given detection time point, an interference detection time window is determined from the sequence of pose difference changes to detect the amplitude of electromagnetic interference. Based on the interference detection time window, the optimal filter is determined for filtering the UAV's pose and the target object's motion data. After this detection time point, the UAV's pose and the target object's motion data are filtered using the optimal filter, and the flight parameters are controlled using the MPC algorithm. At the next detection time point after this detection time point, the interference detection time window is re-determined based on the sequence of pose difference changes, and the time step count of the MPC algorithm is updated using the increase in the window length of the two most recently obtained interference detection time windows.

2. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 1, characterized in that, The specific steps involved in updating the MPC algorithm's time steps are as follows: The increase in window length of the two most recently obtained interference detection time windows is denoted as d1, and the absolute value of the mean of the motion trend amplitude of the two most recently obtained interference detection time windows is denoted as d2. When d1 is greater than or equal to 0, the updated time step is positively correlated with d1 and d2; when d1 is less than 0, the updated time step is set to the initial value of the time step.

3. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 2, characterized in that, The specific steps for obtaining the interference detection time window are as follows: D1: The pose differences at all times within the historical time period before any detection time point constitute a change sequence, and any motion component in the change sequence is denoted as the target sequence; the target sequence is normalized and a value K0 is initialized at the same time; D2: The last K0 elements of the linearly normalized target sequence are denoted as the first group of elements. The mean and standard deviation of the first group of elements are denoted as U1 and A1, respectively. The K0 elements before the first group of elements in the target sequence are denoted as the second group of elements. The mean of the second group of elements is denoted as U2, and the mean of the K0 elements before the second group of elements in the target sequence is denoted as U3. (|U1-U2|+|U2-U3|) / 2 is denoted as the first change magnitude of the target sequence, and |U1-U3| is denoted as the second change magnitude. || represents the absolute value. D3: When both the first and second change amplitudes are greater than A1, the time span of K0 elements is used as an interference detection time window; when the first or second change amplitude is less than or equal to A1, the value of K0 is increased, and D2 and D3 are repeated until an interference detection time window is obtained and then the process stops.

4. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 3, characterized in that, The method for obtaining the motion trend amplitude of the interference detection time window is as follows: During the execution of D3, once an interference detection time window is obtained, U1-U3 is recorded as the motion trend amplitude corresponding to the interference detection time window.

5. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 1, characterized in that, The specific steps involved in determining the optimal filter for filtering UAV pose and target object motion data based on the interference detection time window are as follows: Each component sequence of UAV pose or target object motion data within a historical time period is assigned a filter; the filtered pose difference of the target object relative to the charged wire is obtained by filtering each component sequence using each filter; All filter pose differences within the historical time period constitute the filter change sequence; any motion component in the filter change sequence is denoted as the target filter sequence, and the standard deviation of the elements within the interference detection time window in the target filter sequence is denoted as the electromagnetic interference amplitude after filtering; the mean of the electromagnetic interference amplitudes corresponding to all motion components in the filter change sequence is denoted as the comprehensive interference amplitude after filtering. The filter corresponding to each component sequence with the smallest overall interference amplitude after filtering is taken as the optimal filter.

6. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 5, characterized in that, The specific steps involved in filtering the pose difference of the target object relative to the charged conductor are as follows: All component sequences include: the position in the UAV pose within the historical time period constitutes the first position component sequence, the attitude angle in the UAV pose constitutes the first attitude angle component sequence, the linear acceleration in the target object motion data constitutes the first acceleration component sequence, and the attitude angle in the target object motion data constitutes the second attitude angle component sequence. By sequentially filtering each component sequence using the filters corresponding to all component sequences, we obtain each filtered pose of the UAV and each filtered motion data of the target object. Based on the filtered pose of the UAV and the filtered motion data of the target object, we determine the difference in filtered pose of the target object relative to the charged wire.

7. The anti-interference control method for UAV hoisting operations in confined spaces according to claim 5, characterized in that, The specific steps involved in selecting the filters corresponding to all component sequences that minimize the overall interference amplitude after filtering as the optimal filters are as follows: Each component sequence corresponds to a low-pass filter. A set of candidate values ​​is preset for the cutoff frequency of each low-pass filter. Each low-pass filter selects one of the candidate values. The candidate values ​​selected by the low-pass filters corresponding to all component sequences constitute a combination result. All low-pass filters filter each component sequence based on each combination result to obtain a filtered comprehensive interference amplitude. Traversing all combination results, the combination result with the minimum filtered comprehensive interference amplitude is taken as the optimal parameter of the low-pass filters corresponding to all component sequences. All low-pass filters under the optimal parameter form the optimal filter.

8. The anti-disturbance control method for UAV hoisting operations in confined spaces according to claim 6, characterized in that, The steps of obtaining the pose difference of the target object relative to the charged conductor and determining the filtered pose difference of the target object relative to the charged conductor are both implemented in the following ways: Based on the point cloud on the charged conductor collected in real time by the UAV, a reference coordinate system is established with the charged conductor as the origin; the UAV pose or the filtered pose of the UAV includes: the position and attitude angle of the UAV in the reference coordinate system; the target object motion data or the filtered motion data of the target object includes: the linear acceleration and attitude angle of the target object. The position of the target object relative to the drone is determined based on the linear acceleration and the cycloid length when the target object is hoisted on the drone. The position of the target object relative to the drone and the position of the drone in the reference coordinate system are used to determine the position of the target object relative to the live wire. Based on the attitude angle of the target object and the attitude angle of the UAV in the reference coordinate system, the attitude angle of the target object relative to the charged wire is determined. The position and orientation angle of the target object relative to the charged conductor are denoted as the pose of the target object relative to the charged conductor, and are used as the pose difference or the filtered pose difference.

9. The anti-disturbance control method for UAV hoisting operations in confined spaces according to claim 3, characterized in that, Any motion component in the changing sequence is denoted as the target sequence, and the specific steps included are as follows: Each element of the variation sequence contains two motion components: position and attitude angle. The position of each element is taken as the magnitude and recorded as the position difference of each element. The attitude angle of each element is taken as the magnitude and recorded as the attitude difference of each element. The position differences of all elements in the variation sequence constitute a position difference sequence, and the attitude differences of all elements constitute an attitude difference sequence. Any one of the position difference sequence and the attitude difference sequence is represented as the target sequence.

10. The anti-interference control method for UAV hoisting operations in narrow scenarios according to claim 8, characterized in that, The specific steps involved in controlling flight parameters based on UAV pose and target object motion data, using the MPC algorithm, are as follows: The pose of the target object relative to the charged conductor is determined based on the pose of the UAV and the motion data of the target object. The pose of the target object relative to the charged conductor is used as the state vector of the MPC algorithm, and the three-dimensional linear velocity and angular velocity of the UAV around the direction of gravity are used as flight parameters.

Images