A low-altitude multi-source heterogeneous unmanned aerial vehicle route correction method and device

Abstract

The application discloses a low-altitude multi-source heterogeneous unmanned aerial vehicle route correction method and device, and belongs to the technical field of unmanned aerial vehicle route correction, and comprises the following steps: collecting sensor data, processing the sensor data to output the three-dimensional position of the unmanned aerial vehicle; generating an optimal path dynamically based on the three-dimensional position; determining a control amount based on the optimal path, and outputting a control instruction according to the control amount; the low-altitude multi-source heterogeneous unmanned aerial vehicle route correction method and device reduce the positioning error through multi-source data fusion, meet the high-precision operation requirements such as power inspection, and have enhanced anti-interference capability; the mixed optimization algorithm shortens the obstacle avoidance response time, improves the operation safety and efficiency, and improves the obstacle avoidance success rate; through inter-aircraft communication and a cooperative algorithm, the inter-aircraft collision risk is reduced, the cluster operation efficiency is improved, and the energy consumption is reduced; the multi-source data fusion and the dynamic algorithm adapt to complex electromagnetic environments and weather changes, and improve the operation reliability.

Description

Technical Field

[0001] This invention belongs to the field of UAV flight path correction technology, specifically relating to a method and device for flight path correction of low-altitude multi-source heterogeneous UAVs. Background Technology

[0002] Current drone flight path correction technology suffers from three major pain points: Insufficient positioning accuracy: Traditional single-source GPS / BeiDou positioning has an error of 3-5 meters in scenarios such as electromagnetic interference and urban canyons, which is difficult to meet the high-precision operation requirements of power line inspection and agricultural spraying. For example, a drone-based positioning correction method and system disclosed in publication number CN120408250A achieves correction by clustering historical offset vectors of utility poles, but it does not solve the problem of real-time high-precision positioning in dynamic environments.

[0003] Slow dynamic response: Existing algorithms take more than 20 seconds to replan when dealing with dynamic scenarios such as sudden obstacles and temporary no-fly zones, resulting in an obstacle avoidance failure rate as high as 25%. For example, the real-time target object tracking and positioning method and device with application number 2020107578164 and the collaborative perception positioning method and system with application number 2021115286374 achieve sub-meter level correction through the fusion of laser ranging and image recognition, but do not involve a fast response mechanism in cluster collaborative scenarios.

[0004] Cluster collaboration is inefficient: In multi-machine operations, traditional path planning algorithms can easily increase the risk of collisions between machines by 30%, and it is difficult to balance energy consumption optimization and path efficiency. Research results from institutions such as Chongqing University of Posts and Telecommunications and the China Academy of Information and Communications Technology have adopted multi-model fusion positioning algorithms in their research on multi-mode fusion positioning technology, but they have not solved the collaborative optimization problem at the cluster level.

[0005] To address the aforementioned issues, the following methods are currently the primary approaches used in the field of UAV flight path correction:

[0006] For single-source positioning, the historical offset vector clustering method of utility poles is used. The positioning error is about 1.5 meters in static scenes, but it cannot cope with dynamic scenes such as urban canyons and electromagnetic interference.

[0007] In the dual-mode fusion approach, sub-meter-level correction is achieved by fusing laser ranging and image recognition, with a positioning accuracy of 0.8 meters in power line inspection scenarios. However, the risk of collision between cluster operations increases by 30%.

[0008] In the multi-model algorithm direction, a multi-model fusion positioning algorithm is adopted, which improves the path planning efficiency by 20% in complex terrain, but the dynamic obstacle avoidance response time still requires 15 seconds.

[0009] Furthermore, single-source positioning cannot cope with dynamic scenarios such as urban canyons and electromagnetic interference, dual-mode fusion increases the risk of collisions during swarm operations, and multi-model algorithms have long dynamic obstacle avoidance response times. To address these shortcomings, a new method and device for flight path correction of low-altitude multi-source heterogeneous UAVs needs to be developed to solve the existing problems. Summary of the Invention

[0010] The purpose of this invention is to provide a method and device for correcting flight paths of low-altitude multi-source heterogeneous UAVs, in order to solve the problems of weak perception capabilities, inflexible path planning, and untimely control response in complex low-altitude environments.

[0011] Firstly, this application provides a method for flight path correction of low-altitude multi-source heterogeneous unmanned aerial vehicles (UAVs), including:

[0012] Collect sensor data, process the sensor data and output the three-dimensional pose of the UAV;

[0013] Based on the three-dimensional pose, the optimal path is dynamically generated.

[0014] Based on the optimal path, the control quantity is determined, and control commands are output according to the control quantity.

[0015] In one implementation of the first aspect, the method further includes:

[0016] The system acquires the actual motion state of the UAV after executing control commands, processes the sensor data again to output the UAV's three-dimensional pose, dynamically generates the optimal path, determines the control quantity based on the optimal path, and outputs control commands based on the control quantity.

[0017] In one implementation of the first aspect, the steps of acquiring sensor data, processing the sensor data, and outputting the three-dimensional pose of the UAV include:

[0018] Collect and preprocess sensor data to form a three-dimensional environmental perception network;

[0019] The collected multi-source data is fused using Kalman filtering to obtain observations and state estimates.

[0020] Correct the observations, state estimates, and position components in the Kalman filter.

[0021] In one implementation of the first aspect, dynamically generating the optimal path based on the three-dimensional pose includes:

[0022] Multi-objective cost assessment;

[0023] Ant colony algorithm-based path searching;

[0024] Local paths are optimized based on the Aurora Optimization Algorithm.

[0025] In one implementation of the first aspect, determining the control quantity and outputting a control command based on the control quantity includes:

[0026] The control variables are determined based on the expected position, heading, and state estimate of the optimal path output.

[0027] The control quantity outputs control commands to the actuator.

[0028] In one implementation of the first aspect, optimizing the local path based on the aurora optimization algorithm includes: adjusting energy consumption by dynamically adjusting flight altitude and speed.

[0029] ;

[0030] in, This indicates the estimated total energy consumption. Indicates the start time of the path. Indicates the end time of the path. Indicates the time on the path The estimated speed, Indicates the time on the path The estimated height, The power function is related to speed. Represents a highly correlated power function. This indicates the discretization time step.

[0031] In one implementation of the first aspect, the actuator includes: a main controller and a backup controller for receiving control commands, a main servo motor electric brake, and a backup servo motor electric brake.

[0032] In one implementation of the first aspect, the multi-objective cost evaluation includes:

[0033] The comprehensive cost function value of the current path is determined based on the estimated path length from the current node to the target point, the energy consumption estimation function, and the safety risk function.

[0034] In one implementation of the first aspect, the ant colony search method includes:

[0035] Based on pheromone concentration, pheromone evaporation coefficient, and iterative pheromone increment, the ant colony optimization pheromone update rule is determined.

[0036] Secondly, this application provides a flight path correction device for low-altitude multi-source heterogeneous unmanned aerial vehicles, comprising:

[0037] The processing module is used to collect sensor data, process the sensor data, and output the three-dimensional pose of the UAV.

[0038] The optimal path generation module is used to dynamically generate the optimal path based on the three-dimensional pose.

[0039] The control command output module is used to determine the control quantity based on the optimal path and output control commands according to the control quantity.

[0040] According to the above-mentioned technical solution proposed in this application, the following technical effects can be achieved: This method and device for flight path correction of low-altitude multi-source heterogeneous UAVs reduces positioning errors through multi-source data fusion, meets the requirements of high-precision operations such as power line inspection, and enhances anti-interference capabilities; the hybrid optimization algorithm shortens obstacle avoidance response time, improves operational safety and efficiency, and increases obstacle avoidance success rate; through inter-machine communication and collaborative algorithms, the risk of inter-machine collisions is reduced, the efficiency of cluster operations is improved, and energy consumption is reduced; multi-source data fusion and dynamic algorithms adapt to complex electromagnetic environments and weather changes, improving operational reliability. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below only involve some embodiments of this disclosure and are not a limitation of this disclosure. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the air-to-ground two-level fusion architecture and data flow path according to an embodiment of this application; Figure 2 This is a flowchart of the dynamic path planning algorithm of the method in the embodiments of this application; Figure 3 This is a schematic diagram of the adaptive correction mechanism of the method in the embodiments of this application; Figure 4 This is a risk probability heatmap for the method in the embodiments of this application. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Unless otherwise defined, the technical or scientific terms used in this disclosure shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships, which may change accordingly when the absolute position of the described objects changes. It should be understood that the various steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0045] In a first aspect, according to one or more embodiments, such as Figure 1 , Figure 3 As shown, a method for flight path correction for low-altitude multi-source heterogeneous UAVs is provided, including:

[0046] The airborne perception layer integrates radar, RID, ADS-B, and optoelectronics for all-round environmental perception and self-state acquisition.

[0047] Execution mechanism: As a data hub, it enables communication with the ground and friendly aircraft;

[0048] Ground data processing center: Models and corrects the historical positions of global static reference objects (such as utility poles) to provide reliable global benchmarks for UAVs;

[0049] 5G communication nodes: Building a high-speed, low-latency communication backbone network;

[0050] Uplink data: The drone uploads sensing data and status information to the ground;

[0051] Downlink command: The ground will send high-precision reference information and global planning commands to the UAV;

[0052] Step 1: Multi-source heterogeneous data fusion and high-precision positioning: Receive and process raw data from multiple airborne sensors, and output the UAV's high-precision 3D pose and environmental model including dynamic and static obstacles through Kalman filtering and clustering correction algorithms;

[0053] Kalman filtering and clustering correction together construct a high-precision and high-reliability spatiotemporal reference, providing an accurate environmental map and the UAV's own pose state for subsequent path planning and motion control. Without this step, subsequent planning and control would be meaningless if they were carried out on an incorrect model.

[0054] This step is achieved through the following sub-steps: Step 1-1, Aerial multi-source data acquisition and preprocessing: Integrate LiDAR, visual sensor (photoelectric), millimeter-wave radar, and IMU inertial navigation unit to form a three-dimensional environmental perception network. Simultaneously acquire the three-dimensional point cloud of LiDAR, the image of the visual sensor, the target distance and velocity of the millimeter-wave radar, the angular velocity and acceleration of the inertial measurement unit, and the information of nearby cooperative targets received through RID (Remote Sensing Identification Device) and ADS-B (Automatic Dependent Surveillance-Broadcast).

[0055] Step 1-2, Fusion Estimation Based on Kalman Filtering: Sensor data fusion is achieved through Kalman filtering.

[0056] ;in, : No. The state estimation vector at each sampling time. : No. The state estimation vector at each sampling time point is used to fully describe the motion of the UAV in three-dimensional space. = This includes three-dimensional position, three-dimensional velocity, and roll angle. Pitch angle Yaw angle ; Indicates the first Each sampling time; It is the transpose symbol; The X-axis coordinate of the UAV in three-dimensional space represents its eastward position or a projection of the longitude direction defined by the mission. : The Y-axis coordinate of the UAV in three-dimensional space; representing the northward position or the projection of the latitude direction as defined by the mission; Z-axis coordinate of the drone in three-dimensional space; represents the drone's altitude, not its altitude relative to the takeoff point; : The velocity component of the drone along the X-axis; a positive value indicates flight in the positive X-axis direction. : The velocity component of the drone along the Y-axis; a positive value indicates flight in the positive Y-axis direction. The velocity component of the drone along the Z-axis, i.e., the vertical velocity; a positive value indicates ascent, and a negative value indicates descent. Roll angle refers to the angle of rotation of the drone around its own X-axis, indicating the direction of the nose; pitch angle... The angle at which the drone rotates around its own Y-axis (in the direction of the wings) determines whether the drone dives forward or climbs upward. Yaw / Head: The angle at which the UAV rotates around its own Z-axis (vertical axis), with the geographic North Pole as the reference point (0°) and clockwise as positive; : No. Observation vectors at each sampling time: Source: Multi-source sensors: Provided by lidar ( , , Point cloud; visual provision ( , Pixel and depth information; millimeter-wave radar provides radial range and velocity; IMU provides angular velocity and acceleration; The observation matrix is ​​used to map the state space to the observation space to fuse data from different geometric relationships, such as radar and vision.

[0057] The classic Kalman filter was applied to the fusion of a specific heterogeneous sensor set including radar, RID, ADS-B, optoelectronic, IMU, and RTK, and the corresponding three-dimensional state vector was constructed. and observation matrix This enables high-precision and high-reliability pose estimation that is resistant to interference in complex low-altitude environments;

[0058] Steps 1-3, Cluster-Based Ground Reference Correction: Deploy RTK (Real-Time Kinematic) differential base stations and 5G communication nodes to provide centimeter-level positioning references. Utilize the centimeter-level references provided by the ground RTK differential base stations to systematically correct errors in the Kalman filter output.

[0059] ;

[0060] in, : No. Historical observation offset vectors of reference objects (such as utility poles, building corners), ,in, Observations from airborne sensors From high-precision maps; The first one currently observed by the airborne sensors For the coordinates of ground reference points (such as utility poles or building corners), current technologies mostly use statistical values ​​from historical observation sequences. The known coordinates of this ground reference point in a high-precision map are often obtained by using observations from different times for differential calculation in existing technologies. The number of reference objects of the same type within the current sensing range; existing technologies use a pre-set set of fixed reference stations. The corrected systematic offset vector reflects the common localization error of the current local environment (such as drift caused by RTK signal occlusion); the output of this formula It is not used as a direct control variable, but rather to correct the observations in the Kalman filter. or state estimation Especially correcting the positional components ( , , This eliminates systematic biases and achieves sub-meter level absolute positioning, a crucial step in the transition from relative to absolute positioning.

[0061] Step 2: Dynamic path planning based on hybrid optimization algorithm: using the environment model output from Step 1 (including a static obstacle set, a dynamic target set, a 3D occupancy grid, and a risk heatmap) and a 3D state vector. As input, the CNN-BiLSTM intelligent decision engine first performs situational awareness and high-level strategy generation, performs qualitative analysis of the current environment and situation, and outputs high-level decision instructions (such as emergency obstacle avoidance mode and energy-saving cruise mode) and key decision parameters (such as safety weight adjustment coefficient and search space constraints), thereby guiding the downstream optimization algorithm to converge in a more reasonable direction. The innovative MOWAA-ACO-PLO hybrid optimization algorithm is adopted to dynamically plan or replan an optimal path that meets multiple objectives such as safety, energy saving and efficiency during flight.

[0062] Intelligent Decision Engine: Based on the CNN-BiLSTM deep learning model, it is the brain of the CNN-BiLSTM intelligent decision engine, performing real-time risk assessment, dynamic path decision-making, and generating correction instructions. CNN-BiLSTM is a mature architecture in the field of deep learning. Model structure: ;in, Output vector Output dimension (risk level number or control instruction dimension). : Weights of fully connected layers : Number of hidden units : The feature vector after bidirectional concatenation Bias term, BiLSTM branch (Bidirectional Long Short-Term Memory Network) is a sub-module in the intelligent decision engine that processes time series data. Forward LSTM captures the evolution pattern from history to the present, and backward LSTM captures the implicit trend from the present to the future, identifying the target's motion status (e.g., whether it is approaching at a constant speed, accelerating, or moving away), providing dynamic basis for risk prediction.

[0063] The CNN branch (Convolutional Neural Network) is a submodule in the intelligent decision engine that processes spatial structure data. It is responsible for processing two-dimensional spatial information. The input is a rasterized environment map (such as a 128×128 pixel top view). Each pixel encodes the risk-related attributes of that location (such as obstacle type, no-fly zone boundary, terrain height, etc.). Through multi-layer convolution and pooling, the CNN branch can extract local spatial features (such as a group of buildings 100 meters ahead and open land 50 meters to the right), providing spatial context for risk prediction.

[0064] The MOWAA-ACO-PLO hybrid algorithm makes intelligent decisions based on this benchmark. MOWAA is responsible for multi-objective quantitative evaluation of which flight path is safer and more energy-efficient. ACO, based on the evaluation results of MOWAA, quickly searches for several potential candidate paths in a broad solution space. PLO then refines these candidate paths and finally outputs an optimal flight path to the control layer. It is a collaborative process of qualitative evaluation -> coarse-grained search -> fine-grained optimization.

[0065] Based on the high-precision pose and environment model output in step 1, the optimal path is dynamically generated through the following collaborative sub-steps:

[0066] Step 2-1, Multi-objective cost evaluation based on MOWAA: A hybrid optimization algorithm of MOWAA+ACO+PLO is adopted, and the mathematical model is as follows:

[0067] The above formula is a multi-objective weighted average algorithm. The estimated path length from the current node n to the target point is calculated based on the Euclidean distance in three-dimensional space. Energy consumption prediction function The discretized approximation, where velocity and height Predicted from path shape; Let be the speed-energy function, representing the power required to overcome air resistance and maintain speed. This indicates the speed of the drone relative to the air. The altitude energy consumption function represents the power required to maintain altitude and overcome gravity. Indicates altitude relative to the takeoff point or ground level. The time step represents the flight time for each path segment;

[0068] : Security risk function; ,in It is the three-dimensional distance from path node n to the i-th obstacle (from the environment model). The safety threshold is determined by the results of 3D environment perception. , Energy consumption weight and safety weight are important considerations during emergency obstacle avoidance. Automatically increases; during cruise, Increase the value to optimize energy consumption; f(n) is the core heuristic information for ants choosing paths in the ACO algorithm. , , The computational depth integrates 3D geometry, physical models, and real-time environmental information.

[0069] Energy consumption prediction function The physical modeling, for the first time, integrates the velocity energy consumption function P(v) and the height energy consumption function. The integral form of energy consumption is introduced into the multi-objective cost assessment of path planning, which enables accurate prediction of energy consumption.

[0070] Security risk function Nonlinear modeling, using In this form, the risk value increases exponentially as the safety threshold approaches, which better reflects the urgency of obstacle avoidance than the traditional linear distance penalty.

[0071] Weight adaptive mechanism , It is not a fixed value, but dynamically adjusted according to the real-time risk level, so that the algorithm can automatically switch the optimization focus in different flight scenarios;

[0072] Step 2-2, Swarm intelligent path search based on Ant Colony Optimization (ACO):

[0073] Ant colony optimization pheromone update rules:

[0074] ;

[0075] The pheromone concentration on path edge (i,j) represents the historical excellence of that path segment;

[0076] i: The current node (or the starting point of the edge);

[0077] j: the next node (or the end point of the edge);

[0078] t: At the t-th iteration, the pheromone concentration on the path edge (i,j) is ;

[0079] : Pheromones volatile coefficient, 0 < <1, used to prevent the ant colony optimization algorithm from converging too early;

[0080] The pheromone increment in this iteration is usually proportional to the path quality; the path quality is directly evaluated by the reciprocal of MOWAA's cost function f(n), i.e. ∝ The multi-objective evaluation results of MOWAA are learned and reinforced in the swarm search through the pheromone mechanism of ACO. f(n) is the function value of the multi-objective cost of the current path, which is calculated by the MOWAA algorithm.

[0081] By deeply coupling the MOWAA multi-objective evaluation system with the ACO population search mechanism, the search is guided to quickly converge to the optimal path that satisfies safety, energy saving, and high efficiency.

[0082] Existing ant colony algorithms typically only use path geometry length as the basis for pheromone updates. Introducing the multi-objective cost function calculated by the MOWAA algorithm into the ACO pheromone update mechanism increases the pheromone increment. It is linked to the comprehensive performance of path safety, energy consumption and distance, realizing the deep coupling of multi-objective evaluation and swarm intelligence search, ensuring that the search direction of ACO always points to the comprehensive optimal region defined by MOWAA, and providing a high-quality initial solution for the subsequent fine optimization of PLO.

[0083] Steps 2-3: Fine-tuning of local paths based on the Probability of Loop (PLO) algorithm:

[0084] Particle position update formula:

[0085] Position update equation: ;

[0086] Velocity update equation: Where, a: the index of the particle;

[0087] : The position vector of particle a at the t-th iteration, encoded as a complete candidate flight path;

[0088] : The velocity vector of particle a at the t-th iteration represents the direction and magnitude of the path adjustment, i.e., the correction amount for each waypoint on the current path. Dimensions and same;

[0089] w is the inertia weight. , For acceleration coefficient, , It is a random number;

[0090] x a The position of particle a represents the encoding of a candidate path;

[0091] va The velocity of particle a represents the direction and magnitude of path adjustment.

[0092] , These represent the individual's historical best position and the group's historical best position, respectively.

[0093] The initialization or initial population of each iteration comes from the optimal path found by ACO in the current iteration. PLO does not search randomly, but performs fine development and perturbation near the high-quality solution of ACO to prevent ACO from converging too early.

[0094] Positioning PLO (or a similar optimizer) as a local optimizer or perturbator of ACO forms a two-stage collaborative optimization architecture of ACO global coarse search + PLO local fine-tuning, which improves the robustness and solution quality of the overall ant colony optimization algorithm.

[0095] Energy optimization strategy: Reduce energy consumption while ensuring safety by dynamically adjusting flight altitude and speed.

[0096] ;in, Estimated total energy consumption The estimated speed at time t on the path. Indicates the start time of the path. Indicates the end time of the path. The estimated altitude at time t on the path. Speed-dependent power function : Highly correlated power function Discretization time step, energy consumption is speed function and height function The points.

[0097] Integrating energy optimization strategies into the decision-making process of dynamic path planning allows the ant colony optimization algorithm to directly consider energy consumption during the planning phase.

[0098] After evaluation by MOWAA, search by ACO, and optimization by PLO, the final output is the optimal three-dimensional path that meets the requirements of multiple objectives. This path represents the ideal target trajectory that the UAV needs to fly for a period of time and serves as the expected path for subsequent control modules.

[0099] Step 3: Motion control based on adaptive and dual redundancy mechanism: Using the desired path output in Step 2 as the target value, the adaptive PID control algorithm with online parameter self-tuning is used to calculate the control quantity, and the UAV is driven by dual redundant actuators (servo motor / electric brake) to make its actual motion trajectory accurately track the desired path.

[0100] The adaptive PID control algorithm and dual-redundant control module are responsible for accurately and reliably translating the optimal path command issued by the intelligent decision engine into the physical actions of the UAV; the adaptive PID ensures accurate tracking under various disturbances; and the dual-redundant system serves as the last line of defense, ensuring seamless takeover and continued execution of commands when the main execution channel fails.

[0101] This step is responsible for converting the desired path planned in step 2 into precise and reliable physical actions;

[0102] Step 3-1, Tracking control based on adaptive PID: Solving the problem of accurate trajectory tracking for dynamic systems such as UAVs in complex environments. Formula:

[0103] Where u(t) represents the control quantity, and e(t) represents the tracking error. ,in: , , These represent the desired positions of the UAV in three-dimensional space along the three coordinate axes: : Desired X-axis position (corresponding to eastward or forward position). : Desired Y-axis position (corresponding to north or lateral position). : Desired Z-axis position (corresponding height) This represents the desired course, and t represents the tth iteration. Indicates the course;

[0104] , , Adaptive PID parameters Represents the current flight status (such as speed, load, attitude), with proportional, integral, and differential coefficients adjusted online using the least squares method;

[0105] Adaptive: The coefficients are not fixed and are tuned online using the least squares method. The adaptive PID tracking control continuously monitors the error e(t) and the control quantity u(t). When performance indicators (such as error integral) deteriorate, the control model of the UAV is automatically refitted and updated. , , To adapt to different flight conditions (such as changes in load, wind disturbance); the purpose of refitting is that when flight conditions change (such as increased load, encountering strong wind disturbance, changes in flight altitude / speed), the dynamic characteristics of the UAV will change accordingly. If the PID parameters are still tuned according to the original model, the control performance will decrease or even become unstable. Therefore, it is necessary to re-identify (fit) the current model parameters online based on the real-time collected input and output data, and then calculate the appropriate PID parameters based on the new model.

[0106] Step 3-2, Reliable execution based on dual-redundancy architecture;

[0107] Dual Backup System: Includes two independent controllers (primary and backup) and two actuators (primary servo / electric brake, backup servo / electric brake); the specific structure of the dual redundancy control system adopts a three-layer redundancy architecture of control unit redundancy + actuator redundancy + monitoring and switching unit, ensuring continuous and reliable operation of the flight control system even in the event of a single point of failure; Triple Physical Isolation: the primary controller, backup controller, and monitoring unit are independently powered, have independent PCB layouts, and independent housings, ensuring that a single physical failure (such as circuit burnout or component damage) will not affect other units; Hot Backup + Instantaneous Switching: the backup controller operates in parallel with the primary controller, always maintaining... Synchronization status; switching time is controlled within 10ms, which is much smaller than the attitude response time constant (usually >100ms) for UAV attitude control, achieving seamless takeover; clutch isolation mechanism: the backup servo adopts a normally open clutch design, which is mechanically disengaged from the control surface during normal operation, neither increasing the load on the main system nor affecting the main system when the backup system fails; instantaneous engagement in case of failure, ensuring rapid takeover; multi-dimensional fault detection: the monitoring unit simultaneously monitors control signals, position feedback, current and other multi-dimensional parameters, and can identify various fault modes such as servo jamming, sensor failure, and wire breakage, providing comprehensive fault diagnosis capabilities;

[0108] Transparent upper-level interface: The switching process is completely transparent to the upper-level planning layer of the flight control system. There is no sudden change in the feedback signal (actual flight state) received by the flight control system before and after the switch, ensuring that the upper-level algorithm does not need to be aware of the lower-level switch, which simplifies the system design.

[0109] The primary and backup channels receive the same control quantity u(t), but only the primary channel outputs to drive the aircraft. A highly reliable monitoring and switching unit continuously compares the execution feedback of the primary channel with the expected result. Once a failure is detected (such as servo jamming or response timeout), the primary channel is cut off and the backup channel is connected within milliseconds to achieve seamless takeover.

[0110] By combining adaptive PID with online parameter self-tuning algorithms and a hardware dual-redundancy architecture with fast switching logic, the adaptability and reliability of the control system in complex environments are improved, achieving a fast dynamic response and high reliability as a fundamental guarantee.

[0111] It is equipped with two completely independent controllers and actuators (main servo motor / electric brake, backup servo motor / electric brake); the main and backup channels receive the same control quantity u(t), but only the main channel outputs. An independent high-speed monitoring unit compares the feedback and commands of the main channel in real time.

[0112] Switching logic: Once the monitoring unit detects a failure in the main channel (such as servo jamming or response timeout), it will automatically disconnect the main channel within milliseconds and seamlessly connect to the backup channel, allowing the standby system to continue executing control commands to ensure flight continuity and safety.

[0113] Step 4: State Feedback and Closed-Loop Optimization. The actual motion state of the UAV after executing control commands (estimated in real time by Step 1) is fed back to Steps 2 and 3 in real time to form closed-loop control, thereby achieving continuous optimization and dynamic adjustment of the entire flight process.

[0114] The execution effect of the control layer (actual flight trajectory) is monitored in real time by the perception layer and fed back to the planning layer. If the actual trajectory deviates from the planned path, or if the environment changes, the planning layer can immediately trigger replanning, forming an intelligent closed loop of perception-planning-control-feedback.

[0115] In this embodiment, the hardware configuration includes: radar, photoelectric, RID, ADS-B, RTK positioning module, 5G communication module, and dual-redundant control module. The ground base station deploys an RTK differential base station, 5G communication nodes, and a data processing center.

[0116] Software system: Develop a dynamic path planning algorithm module and an intelligent decision engine, and integrate them into the UAV flight control system.

[0117] Task planning: The ground station inputs task parameters (start point, end point, obstacle avoidance area), and the dynamic path planning algorithm module automatically generates the initial path.

[0118] Real-time correction: During flight, sensor data is uploaded to the intelligent decision engine in real time, and the intelligent decision engine dynamically adjusts the path and triggers correction commands.

[0119] Cluster collaboration: Multiple machines exchange location and status information through 5G communication to coordinate obstacle avoidance and path optimization.

[0120] Data feedback: After the task is completed, the dynamic path planning algorithm module generates high-precision trajectory data and energy consumption report for subsequent optimization and analysis.

[0121] In regional UAV monitoring scenarios, this invention improves positioning accuracy compared to traditional methods, reduces obstacle avoidance response time to ≤8 seconds, increases obstacle avoidance success rate and swarm operation efficiency, and reduces energy consumption; it also enhances anti-electromagnetic interference capabilities and adapts to the needs of complex environments.

[0122] The intelligent decision engine (CNN-BiLSTM) can predict and characterize sudden intrusion targets that cannot communicate, have unpredictable trajectories, but can be detected by multiple sources of sensors during radar detection, such as kites, balloons, unauthorized drones, and birds.

[0123] Input processing: In At any given moment, the radar point cloud sequence shows that the trajectory of the kite (hypothetical) target continues to move towards the intersection of the flight paths, and the electro-optical image confirms that it is a flexible and difficult-to-penetrate target; the BiLSTM branch identifies the target's approaching situation from the historical sequence.

[0124] Risk prediction: Combining a CNN branch with a restricted area layer, an image of the next 3 seconds is generated. Figure 4 The risk probability heatmap shown illustrates that the risk probability rapidly rises to over 80% at the intersection of the original flight path and the kite's trajectory. The dashed line represents the original planned flight path, the red cross indicates the intersection point between the kite's current position and the predicted trajectory, the green arrow represents the avoidance direction recommended by the intelligent decision engine (CNN-BiLSTM) (right turn and climb), and the light gray dashed line represents the kite's predicted trajectory. The intersection point is located at the junction of the original flight path and the kite's trajectory, with a risk probability as high as 96%. The model suggests a 15° right turn and a 5m / s climb, reducing the risk to below 12%.

[0125] Decision output: Urgency score = 0.9 (extremely high); directly triggered the entire system's emergency obstacle avoidance mode;

[0126] Urgency score ;

[0127] Among them, P risk It predicts the collision probability, and its value in the kite scene is 0.85, which comes from the intersection probability in the CNN branch heatmap.

[0128] T impact It is the predicted collision time (normalized, remaining / warning threshold), with a value of 0.9 (warning threshold), calculated from the BiLSTM branch velocity and distance;

[0129] C severity The value of 0.8 (for flexible targets with a high risk of entanglement) represents the severity of the collision consequences and is derived from photoelectric classification and rule base.

[0130] α, β, γ are weighting coefficients with values ​​of 0.4, 0.4, and 0.2, respectively. These values ​​are derived from experimental calibration and are dynamically increased during emergency obstacle avoidance mode.

[0131] Calculate: E score =0.4×0.85+0.4×0.9+0.2×0.8=0.86, Decision logic: E score>0.8 → Triggers emergency obstacle avoidance mode, 0.5 <E score ≤0.8 → Trigger warning mode (early deceleration, enhanced monitoring), E score ≤0.5 → Normal cruising.

[0132] Quantitative optimization of dynamic path planning (MOWAA+ACO+PLO):

[0133] MOWAA weight dynamic adjustment: Due to receiving an urgency score of 0.9, the security weight is automatically adjusted. The energy consumption weight was increased from 0.4 to 0.8. The corresponding reduction in multi-objective cost. Safety has become the overriding objective.

[0134] ACO Search Guidance: The right-turn climbing suggestions and risk probability heatmaps output by the intelligent decision engine are transformed into strong initial pheromones and heuristic information for the ACO algorithm, guiding ants to prioritize exploring the safe area on the upper right.

[0135] PLO fine-tuning: ACO quickly finds several feasible paths to bypass kites and restricted areas, performs local optimization of path smoothness and energy consumption, and selects a three-dimensional space curve that balances emergency avoidance and subsequent endurance.

[0136] Dual redundancy control and precise execution of adaptive PID:

[0137] Error Calculation: Upon receiving the first desired waypoint of the new flight path, the adaptive PID controller immediately obtains the current actual three-dimensional position and attitude of the UAV from the fusion center, and calculates the six-degree-of-freedom error e(t), for example, the lateral position error e. y Possibly negative (needs to be tilted to the right), height error e z Positive (requires climbing).

[0138] Parameter Adaptation: Due to the emergency high-speed maneuver, both the error e(t) and the rate of change of error de(t) / dt are large. The controller instantaneously increases the differential gain K according to the built-in online tuning rules (such as least squares identification). d To suppress overshoot, the proportional gain K is appropriately increased. p To improve response speed while maintaining integral gain K i Smaller size prevents saturation.

[0139] Dual redundancy execution guarantee:

[0140] Main channel execution: The controller outputs u(t), the main servo system drives the aileron to deflect to the right, the elevator to deflect upward, and at the same time the electric braking system applies slight braking to the left motor to assist steering.

[0141] Real-time monitoring and switching: An independent monitoring unit compares the actual deflection angle fed back by the servo with the commanded deflection angle. Assuming a sudden jamming of the main and aileron servos (feedback angle stall) during a violent maneuver, the redundancy switch detects this fault within 10 milliseconds; it is a core component of the dual-redundancy control module.

[0142] Seamless takeover: The switching logic immediately cuts off the drive signal of the main channel and activates the backup servo system and electric braking system. The backup channel takes over from the control state of the previous moment and continues to execute the right turn and climb correction command. For the UAV flight control system, the switch is transparent and the UAV attitude is not disturbed.

[0143] The energy optimization strategy accompanies the entire process: After the new flight path is stabilized, the energy optimization module starts to work. The energy optimization module belongs to the intelligent flight management computer subsystem in the airborne intelligent perception and control system (UAV end); under the premise of ensuring mission completion and safety, it optimizes energy consumption by dynamically adjusting flight speed, altitude and attitude.

[0144] This embodiment describes how a drone carrying out an emergency mission safely and efficiently completes its flight in a complex low-altitude urban environment.

[0145] Task initialization:

[0146] The drone takes off from point A and is headed to point B. The low-altitude digital twin platform of the ground data processing center plans an initial efficient flight path for it and sends it to the drone. The low-altitude digital twin platform is a core component of the ground reference layer.

[0147] Multi-source heterogeneous sensing and fusion (airborne):

[0148] Radar: Continuously scans ahead to detect physical obstacles such as buildings and power lines, as well as sudden obstacles / intrusion targets (D3, such as kites);

[0149] Optoelectronics: Optical identification and confirmation of suspicious targets (such as kites) detected by radar.

[0150] RID: Continuously broadcasts its own identity, location, speed, and heading, and receives RID information from other nearby cooperating drones (D1) to achieve mutual visibility; D1: Other cooperating drones (networked via RID, capable of communication and collaboration); D2: Manned aircraft (broadcast via ADS-B, capable of detection and avoidance); D3: Sudden obstacles / intrusion targets (such as kites, balloons, or drones flying illegally without communication links, only detectable by radar / electro-optical sensors);

[0151] ADS-B: Receives broadcast signals from overhead general aviation helicopters (D2) to detect potential conflicts in advance.

[0152] IMU / GNSS: Provides its own basic navigation information.

[0153] The data fusion center uses Kalman filtering to fuse the above information into a real-time situational map that includes a static map, dynamic cooperative targets, and dynamic non-cooperative threats. The airborne fusion computing unit belongs to the airborne intelligent perception and control system (UAV end).

[0154] Dynamic route planning and collaborative decision-making (air-ground collaboration):

[0155] Scenario: During flight, the ground control center detects a sudden temporary aerial activity (such as firefighting drone operations) ahead of the flight path via wide-area radar, creating a temporary no-fly zone; at the same time, the airborne radar detects a rapidly approaching non-cooperative flying object (kite) from the side.

[0156] Decision-making process, such as Figure 2 As shown: MOWAA (Multi-Target Assessment): The intelligent decision engine receives threat information; Recalculated at this time (Safe distance) weight The situation is rapidly escalating; safety must be the top priority.

[0157] ACO (Pathfinder Search): Starting from the current location, it quickly generates multiple alternative paths, bypassing temporary no-fly zones and kites, and bases them on historical success rates (pheromones). Evaluate the merits of each path.

[0158] PLO (Dynamic Optimization): Considering the complexity of urban wind fields, the optimal path recommended by ACO is fine-tuned (updating particle position x). a and speed v a To find a new path that is both safe and relatively stable and energy-efficient;

[0159] Air-to-ground synchronization: New dynamically optimized routes are simultaneously uploaded to the ground center for record-keeping and then issued to the drones for execution; the ground center can also intervene forcibly and issue new waypoint instructions.

[0160] Adaptive correction mechanism execution (over-the-air):

[0161] Once the intelligent decision engine determines the new path, it immediately sends a correction command to the dual-redundant control module.

[0162] The adaptive PID controller starts working: it compares the current state (slightly deviated due to wind disturbance) with the desired new trajectory, calculates the tracking error e(t), and generates the control quantity u(t), which drives the torque-redundant fly-by-wire system and vector thruster, enabling the UAV to smoothly and quickly enter the new flight path.

[0163] If the main controller malfunctions due to electromagnetic interference, the redundancy switch will activate the backup controller within milliseconds to ensure uninterrupted flight safety.

[0164] Throughout the process, the Energy Optimization Manager dynamically adjusts the cruising speed based on the new route length and complexity, saving as much power as possible while meeting the timeliness of emergency delivery.

[0165] Cluster collaboration and efficiency:

[0166] Throughout the airspace, multiple drones exchange intentions via RID and a low-altitude private network. When the ground control center replans the path for a drone, it simultaneously notifies other drones in the area to coordinate avoidance and prevent chain reaction congestion.

[0167] Ultimately, the drones completed the entire process of perception, decision-making, and maneuvering within ≤8 seconds, successfully avoiding all dynamic and static threats and delivering supplies safely. Compared to fixed-route solutions, the overall efficiency of the cluster was improved and energy consumption was reduced.

[0168] Secondly, according to one or more embodiments, a flight path correction device for low-altitude multi-source heterogeneous unmanned aerial vehicles is provided, comprising:

[0169] The processing module is used to collect sensor data, process the sensor data, and output the three-dimensional pose of the UAV.

[0170] The optimal path generation module is used to dynamically generate the optimal path based on the three-dimensional pose.

[0171] The control command output module is used to determine the control quantity based on the generated optimal path and output control commands according to the control quantity.

[0172] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0173] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0174] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for flight path correction for low-altitude multi-source heterogeneous unmanned aerial vehicles (UAVs), characterized in that: include: Collect sensor data, process the sensor data, and output the UAV's three-dimensional pose and environmental model of dynamic and static obstacles; Based on the three-dimensional pose and the environment model of dynamic and static obstacles, the optimal path is dynamically generated. Based on the optimal path, the control quantity is determined, and the control command is output according to the control quantity; The dynamic generation of the optimal path based on the environment model of the three-dimensional pose and dynamic and static obstacles includes: Multi-objective cost assessment; Based on the multi-objective evaluation results, the pheromone concentration, pheromone volatility coefficient, and iterative pheromone increment are used to determine the ant colony optimization pheromone update rule and output the initial solution. Based on the initial solution, the local path is optimized using the aurora optimization algorithm.

2. The method for flight path correction for low-altitude multi-source heterogeneous UAVs according to claim 1, characterized in that: The method further includes: The system acquires the actual motion state of the UAV after executing control commands, processes the sensor data again to output the UAV's three-dimensional pose, dynamically generates the optimal path, determines the control quantity based on the optimal path, and outputs control commands based on the control quantity.

3. The method for flight path correction for low-altitude multi-source heterogeneous UAVs according to claim 1, characterized in that: The process of collecting sensor data, processing the sensor data, and outputting the UAV's three-dimensional pose and environmental model of dynamic and static obstacles includes: Collect and preprocess sensor data to form a three-dimensional environmental perception network; The collected multi-source data is fused using Kalman filtering to obtain observations and state estimates. Correct the observations, state estimates, and position components in the Kalman filter.

4. The method for flight path correction for low-altitude multi-source heterogeneous UAVs according to claim 1, characterized in that: The process of determining the control quantity and outputting control commands based on the control quantity includes: The control variables are determined based on the expected position, heading, and state estimate of the optimal path output. The control quantity outputs control commands to the actuator.

5. A method for flight path correction for low-altitude multi-source heterogeneous unmanned aerial vehicles according to claim 1, characterized in that: The local path optimization based on the aurora optimization algorithm includes: adjusting energy consumption by dynamically adjusting flight altitude and speed. ； in, This indicates the estimated total energy consumption. Indicates the time on the path The estimated speed, Indicates the start time of the path. Indicates the end time of the path. Indicates the time on the path The estimated height, Represents the speed-dependent power function. Represents a highly correlated power function. This indicates the discretization time step.

6. The method for flight path correction for low-altitude multi-source heterogeneous UAVs according to claim 4, characterized in that: The actuators include: a main controller and a backup controller for receiving control commands, a main servo motor electric brake, and a backup servo motor electric brake.

7. A method for flight path correction for low-altitude multi-source heterogeneous UAVs according to claim 1, characterized in that: The multi-objective cost assessment includes: The comprehensive cost function value of the current path is determined based on the estimated path length from the current node to the target point, the energy consumption estimation function, and the safety risk function.

8. A flight path correction device for low-altitude multi-source heterogeneous unmanned aerial vehicles, characterized in that: include: The processing module is used to collect sensor data, process the sensor data, and output the three-dimensional pose of the UAV. The optimal path generation module is used to dynamically generate the optimal path based on the three-dimensional pose. The control command output module is used to determine the control quantity based on the optimal path and output control commands according to the control quantity. The optimal path generation module includes: Multi-objective cost evaluation unit, used for multi-objective cost evaluation; The swarm search unit is used to determine the ant colony optimization pheromone update rule based on the multi-objective evaluation results, pheromone concentration, pheromone volatility coefficient and iterative pheromone increment, and output the initial solution; The local path optimization unit is used to optimize the local path based on the initial solution and the aurora optimization algorithm.

Images