Unmanned aerial vehicle cluster formation control method and system based on multi-level intelligent algorithm

Abstract

The application relates to the technical field of unmanned aerial vehicle autonomous control, and discloses a multi-level intelligent algorithm-based unmanned aerial vehicle cluster formation control method and system, which obtains external environment monitoring data and motion state data of an unmanned aerial vehicle cluster, constructs a multi-target comprehensive cost function, performs optimization on an initial trajectory of cluster formation, obtains an optimal trajectory set satisfying a dynamic constraint, performs collision risk evaluation on sampling points on the optimal trajectory set, and outputs a trajectory safety judgment result; a Laplace matrix is used to perform mathematical representation and eigenvalue modulation on a formation topological structure, target formation parameters adapted to environment constraints are generated, and finally, the optimal trajectory set and the target formation parameters are converted into bottom-layer flight control instructions to drive the unmanned aerial vehicle to perform a flight action while collecting real-time motion state data. Therefore, the application solves the problems of slow trajectory optimization convergence, poor real-time collision detection, insufficient formation dynamic adaptation and weak system module cooperation in the prior art.

Description

Technical Field

[0001] This invention relates to the field of autonomous control technology for unmanned aerial vehicles (UAVs), and in particular to a method and system for controlling UAV swarm formation based on multi-level intelligent algorithms. Background Technology

[0002] With the rapid iteration of UAV technology and the explosive growth in demand for swarm collaboration, UAV swarm formation control technology has become a core supporting technology for the efficient utilization of airspace resources and the collaborative execution of complex tasks. However, existing technologies still face many technical bottlenecks in practical applications, making it difficult to adapt to the high-precision, high-stability formation operation requirements in complex dynamic environments. Specifically:

[0003] Limitations of trajectory optimization techniques: Existing techniques mostly employ traditional gradient descent or particle swarm optimization algorithms for trajectory optimization, which suffer from slow convergence speed and a tendency to get trapped in local optima. Especially under complex environments and external disturbances, formations struggle to maintain stable formation structures, resulting in low trajectory optimization efficiency.

[0004] Limitations of collision detection capabilities: Traditional collision detection methods are mainly based on geometric shape detection or grid maps, resulting in high computational complexity and poor real-time performance. In dynamic obstacle environments, existing technologies struggle to simultaneously handle collision detection between the formation and external obstacles, as well as collision detection between drones within the formation, posing safety risks.

[0005] Insufficient formation representation and adaptability: Existing formation control methods lack effective mathematical representation of formations, making them prone to instability or uneven transitions when changing formations in complex environments. This is particularly problematic when navigating narrow passages, where dynamic adaptive adjustment of the formation is impossible.

[0006] Low system integration: Existing technologies often treat trajectory planning, collision detection, and formation control as independent modules, lacking organic integration and making it difficult to achieve global optimization and collaborative control. Summary of the Invention

[0007] This invention provides a method and system for controlling UAV swarm formation based on a multi-level intelligent algorithm, aiming to solve the problems of slow trajectory optimization convergence, poor real-time collision detection, insufficient dynamic adaptation of formation, and weak system module coordination in the prior art.

[0008] To achieve the above objectives, this invention provides a method for controlling UAV swarm formation based on a multi-level intelligent algorithm, the method comprising the following steps:

[0009] The system acquires external environmental monitoring data and motion status data of the UAV swarm, performs multimodal fusion preprocessing on the environmental monitoring data and motion status data, and achieves global sharing of the dataset through the swarm communication network to obtain a synchronized standardized environmental dataset and swarm status dataset.

[0010] Based on the standardized environment dataset and cluster state dataset, a multi-objective comprehensive cost function is constructed, and an improved quasi-Newton method is used to optimize the initial trajectory of the cluster formation to obtain the optimal trajectory set that satisfies the dynamic constraints.

[0011] Using the standardized environmental dataset, a Euclidean symbolic distance field is constructed, and a collision risk assessment is performed on the sampling points on the optimal trajectory set to output the trajectory safety determination result.

[0012] Based on the cluster state dataset and trajectory safety determination results, mathematical representation and eigenvalue modulation of the formation topology are performed using the Laplace matrix to generate target formation parameters that adapt to environmental constraints.

[0013] The optimal trajectory set and target formation parameters are converted into low-level flight control commands to drive the UAV to perform flight maneuvers, while simultaneously collecting real-time motion status data.

[0014] Optionally, the external environment monitoring data includes obstacle distribution point cloud, environmental channel width, real-time wind speed, and communication interference intensity, and the motion state data includes the UAV's pose, speed, and acceleration.

[0015] Optionally, multimodal fusion preprocessing is performed on the environmental monitoring data and motion state data, and global sharing of the dataset is achieved through a cluster communication network to obtain synchronized standardized environmental datasets and cluster state datasets, specifically including:

[0016] The raw environmental monitoring data and raw motion state data are preprocessed by denoising, registration and normalization to obtain standardized environmental datasets and cluster state datasets.

[0017] The standardized environmental dataset and cluster status dataset are transmitted to the decision-making units of all UAVs in the cluster using the established cluster communication network, ensuring data synchronization.

[0018] Optionally, based on the standardized environment dataset and the cluster state dataset, a multi-objective comprehensive cost function is constructed, specifically including:

[0019] Constructing a path length cost Energy consumption cost Security costs and formation stability cost The multi-objective comprehensive cost function is expressed as follows:

[0020]

[0021] In the formula, The preset weighting coefficients are adjusted for adaptability based on different task types.

[0022] Optionally, an improved quasi-Newton method is used to optimize the initial trajectory of the swarm formation to obtain an optimal trajectory set that satisfies dynamic constraints, specifically including:

[0023] Based on the standardized environment dataset, an environment complexity index is calculated, and then an environment complexity factor is derived from the environment complexity index. The specific expression is as follows:

[0024]

[0025]

[0026] In the formula, As the basic weight for environmental indicators, For obstacle distribution complexity, For meteorological interference intensity, For spatial constraint coefficients, Electromagnetic environment index; The basic regularization coefficient, Let the current gradient norm be... As an environmental complexity factor, This is the sensitivity coefficient;

[0027] An identity matrix is ​​introduced to maintain the positive definiteness of the Hessian matrix inverse. An improved BFGS formula is used to update the approximate value of the Hessian matrix inverse. The specific expression is as follows:

[0028] In the formula, For the first The inverse approximation of the Hessian matrix in the next iteration , The gradient change vector, Update the vector for the parameters. It is the identity matrix. Environmental complexity factor;

[0029] The optimization step size is dynamically adjusted based on the environmental complexity index, as expressed in the following expression: In the formula, This is the initial step size;

[0030] The trajectory optimization is completed by combining the adjusted step size with the updated inverse approximation of the Hessian matrix, and the optimal trajectory set is output.

[0031] Optionally, the calculation of environment complexity metrics also includes:

[0032] Computational obstacle distribution complexity The expression is as follows: In the formula, To determine the number of obstacles in the scanned area, For the sensor scan volume, For drones to the first The closest distance to each obstacle The standard deviation of the obstacle distance distribution. , , These are the weighting coefficients;

[0033] Calculate the intensity of meteorological disturbance The expression is as follows: In the formula, For real-time wind speed, For the maximum wind resistance speed of the drone, For wind speed change rate, These are the weighting coefficients;

[0034] Calculate spatial constraint coefficients The expression is as follows: In the formula, To achieve the minimum width of the passable channel, , For the wingspan of the drone, This is a safety margin for height above the ground / top. , These are the weighting coefficients;

[0035] Calculate the electromagnetic environment index The expression is as follows: In the formula, For communication interference signal strength, For wireless communication bit error rate, These are the weighting coefficients;

[0036] right Perform normalization; the specific expression is: Substitute the normalized result into the environmental complexity index formula to complete the calculation.

[0037] Optionally, a Euclidean symbolic distance field is constructed using the standardized environmental dataset, and a collision risk assessment is performed on the sampling points on the optimal trajectory set to output a trajectory safety determination result, specifically including:

[0038] The obstacle point cloud data of the standardized environmental dataset is converted to a resolution of [resolution value missing]. A 3D raster map using heuristic functions Generate the initial path, where, The actual cost from the starting point to node n. To heuristically estimate the cost from n to the target point, a smooth trajectory is generated with the objective of minimizing the time derivative of acceleration. The specific expression is as follows: In the formula, The function represents the three-dimensional spatial trajectory parameters of the UAV, where T represents the complete execution time, t represents the current time, and d represents the preset trajectory smoothing coefficient.

[0039] Construct the Euclidean symbolic distance field for any point in space. The specific expression for calculating the distance value is as follows: In the formula, For the spatial location of obstacles, Characteristic points Positional relationship with obstacles;

[0040] Perform ESDF value detection on each sampling point of the optimal trajectory set. If the ESDF value is less than the safety threshold... If an external collision risk is detected, the minimum distance between drones within the cluster will be checked; if it is less than the safe formation spacing... If an internal collision risk is detected, a trajectory safety assessment result will be output.

[0041] Optionally, based on the cluster state dataset and trajectory safety determination results, mathematical representation and eigenvalue modulation of the formation topology are performed using the Laplace matrix to generate target formation parameters adapted to environmental constraints. Specifically, this includes: constructing an undirected topology graph based on cluster communication connection relationships. ,in For a collection of drone nodes, This is the set of communication edges between nodes;

[0042] According to the topology diagram Calculate the degree matrix Adjacency Matrix This leads to the Laplace matrix, which is expressed as follows:

[0043]

[0044] The eigenvalue decomposition of the Laplacian matrix is ​​expressed as follows:

[0045]

[0046] In the formula, It is an eigenvalue diagonal matrix. The eigenvector matrix;

[0047] use The eigenvectors corresponding to non-zero eigenvalues ​​represent the formation topology, and the eigenvectors corresponding to zero eigenvalues ​​represent the overall translational characteristics of the formation, generating initial formation coding data.

[0048] Based on the environmental constraints in the trajectory safety determination result, non-zero feature values ​​are scaled proportionally to generate the target formation parameters, the specific expression of which is: In the formula, These are the original eigenvalues. Adjust the scaling factor to suit the environment. Optionally, the optimal trajectory set and target formation parameters are converted into low-level flight control commands to drive the UAV to perform flight maneuvers, while simultaneously collecting real-time motion state data, specifically including:

[0049] The optimal trajectory set and target formation parameters are converted into attitude control commands and power drive commands for each UAV.

[0050] The control commands are transmitted to the drone's actuators, which drive the rotor system to complete the corresponding flight maneuvers.

[0051] The pose and velocity of the drone after its actions are collected by the state acquisition device, and the cluster state dataset is updated after preprocessing.

[0052] Furthermore, to achieve the above objectives, the present invention also provides a UAV swarm formation control system based on a multi-level intelligent algorithm. The system includes: a perception layer having an environmental perception module, a state estimation module, and a communication module; a decision layer having a trajectory optimization module, a collision detection module, and a formation representation module; and an execution layer having a flight control module and an actuator module. The perception layer, the decision layer, and the execution layer are configured to jointly execute the UAV swarm formation control method based on the multi-level intelligent algorithm described above.

[0053] The beneficial effects of this invention are as follows:

[0054] (1) Significantly improved trajectory optimization performance: Compared with the traditional gradient descent method, the number of iterations is reduced by adopting the quasi-Newton method; the direct calculation of the Hessian matrix is ​​avoided, which greatly reduces the iteration time; at the same time, environmental complexity is fully introduced, and the adaptability of the quasi-Newton method and trajectory optimization under the influence of environmental complexity are improved.

[0055] (2) Collision detection capability is greatly enhanced: the accuracy is significantly improved compared with traditional geometric detection methods; the ESDF construction and query time is controlled within 5ms; multiple types of collisions are handled in a unified manner, and external obstacles and internal collisions within the formation are handled simultaneously;

[0056] (3) Significantly improved formation adaptability: formation change time is shortened; the use of eigenvalue modulation technology makes the formation change process smoother; and it can pass through narrow channels with a smaller width by forming up.

[0057] (4) The system has excellent overall performance, improved formation stability and mission success rate, and can quickly reconstruct formations in the event of a single machine failure.

[0058] (5) Strong engineering practicality: modular design facilitates maintenance and upgrades; moderate computational complexity is suitable for embedded system deployment; adjustable parameters can adapt to different application scenarios. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the overall technical process of the present invention.

[0060] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0061] Figure 3 This is a schematic diagram illustrating the system execution of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0063] like Figure 1 As shown in this embodiment, a UAV swarm formation control method based on a multi-level intelligent algorithm is proposed, which includes the following steps:

[0064] S1: Acquire external environment monitoring data and motion status data of the UAV cluster, perform multimodal fusion preprocessing on the environmental monitoring data and motion status data, and achieve global sharing of the dataset through the cluster communication network to obtain a synchronized standardized environmental dataset and cluster status dataset.

[0065] It should be noted that the external environment monitoring data includes obstacle distribution point cloud, environmental passage width, real-time wind speed, and communication interference intensity.

[0066] In practical applications, the acquisition of external environment monitoring data relies on a network of multiple types of sensors. Specifically, obstacle distribution point clouds are collected by lidar with a sampling period of 10Hz and a point cloud resolution controlled at 0.05m, ensuring that small obstacles with a diameter of at least 0.1m can be effectively identified. Environmental passage width is measured by millimeter-wave radar with a measurement error not exceeding 0.2m. Real-time wind speed is collected by an ultrasonic anemometer with a sampling accuracy of 0.1m / s, accurately capturing instantaneous wind speed fluctuations. Communication interference intensity is acquired by a radio frequency monitoring module, with the monitoring frequency band covering the dedicated communication frequency band for UAV swarms, ensuring targeted communication status perception.

[0067] It should be noted that the motion state data includes the drone's pose, speed, and acceleration.

[0068] In practical applications, motion state data is acquired using the UAV's onboard sensor system. The Inertial Measurement Unit (IMU) collects triaxial acceleration and angular velocity data at a sampling frequency of 200Hz, providing the foundation for attitude calculation. GPS and a visual positioning module work together to achieve high-precision attitude calculation for the UAV, with a positioning error ≤0.5m in open environments and no more than 1m in partially obstructed environments. The UAV's real-time speed is calculated by combining the motor speed sensor and the electronic speed controller, with a speed measurement error ≤0.1m / s. All acquired data is equipped with a high-precision timestamp, with time synchronization errors controlled within 1ms, ensuring the temporal consistency of multi-source data.

[0069] Specifically, step S1 is executed as follows:

[0070] S11: Denoise, register, and normalize the original environmental monitoring data and the original motion state data to obtain a standardized environmental dataset and a cluster state dataset.

[0071] It should be noted that the raw collected data needs to undergo three preprocessing steps: denoising, registration, and normalization, before it can be formed into a standardized dataset.

[0072] The denoising process employs differentiated strategies for different types of data: statistical filtering is used to remove outliers from LiDAR point clouds, and a threshold for the number of neighboring points is set. When the number of neighboring points of a point is lower than the threshold, it is identified as a noise point and removed. Kalman filtering is used to suppress high-frequency vibration interference for IMU data. Its state equation is constructed based on the rigid body motion model of the UAV, and the observation equation is fused with GPS positioning data to achieve error compensation. Moving average filtering is used for wind speed and communication interference data, with the sliding window length set to 5 sampling periods to ensure data smoothness while taking into account real-time performance.

[0073] The registration process achieves spatial and temporal alignment of multi-source data: spatial registration is completed by using an iterative nearest-point algorithm for LiDAR point clouds and visual positioning data, so that the pose coordinates of environmental obstacles and UAVs are unified to the same coordinate system; linear interpolation is performed on temporal data according to timestamps to fill in data missing caused by differences in sensor sampling frequencies, ensuring the integrity of various types of data at the same time node.

[0074] The normalization process employs the min-max normalization method, mapping all environmental and state data to the [0,1] interval to eliminate the impact of dimensional differences on subsequent algorithms.

[0075] S12: Utilize the established cluster communication network to transmit the standardized environmental dataset and cluster status dataset to the decision-making units of all UAVs within the cluster, ensuring data synchronization.

[0076] It should be noted that data sharing is achieved through a dedicated self-organizing network communication system for UAV swarms. Data transmission adopts a publish-subscribe model. The data preprocessing unit acts as the publisher, pushing standardized datasets to the communication network at fixed intervals. Each UAV decision-making unit acts as a subscriber, synchronously receiving the datasets and updating its local database. The system is equipped with a heartbeat packet verification mechanism. If a data packet is lost, a retransmission process is triggered to ensure data synchronization across all nodes in the swarm, providing a unified data foundation for distributed decision-making.

[0077] S2: Based on the standardized environment dataset and cluster state dataset, a multi-objective comprehensive cost function is constructed, and the improved quasi-Newton method is used to optimize the initial trajectory of the cluster formation to obtain the optimal trajectory set that satisfies the dynamic constraints.

[0078] Specifically, step S2 is executed as follows:

[0079] S21: Construct a system that includes path length cost Energy consumption cost Security costs and formation stability cost The multi-objective comprehensive cost function is expressed as follows:

[0080]

[0081] In the formula, The preset weighting coefficients are adjusted for adaptability based on different task types.

[0082] In this embodiment, to balance the economy, security, and formation stability of the trajectory, a path length cost is included. Energy consumption cost Security costs and formation stability cost The multi-objective comprehensive cost function. The weighting coefficients can be adapted according to the task type.

[0083] S22: Calculate the environmental complexity index based on the standardized environmental dataset, and then derive the environmental complexity factor through the environmental complexity index. The specific expression is as follows:

[0084]

[0085]

[0086] In the formula, The basic weights for environmental indicators (default 0.35, 0.25, 0.25, 0.15). For obstacle distribution complexity, For meteorological interference intensity, For spatial constraint coefficients, Electromagnetic environment index; The basic regularization coefficient, Let the current gradient norm be... As an environmental complexity factor, This is the sensitivity coefficient.

[0087] In practical applications, the basic weights of environmental indicators can be adjusted for different tasks, as shown in Table 1:

[0088] Task type <![CDATA[C obs ]]> <![CDATA[C wea ]]> <![CDATA[C spa ]]> <![CDATA[C em ]]> Urban logistics distribution 1.2 1.0 1.5 1.3 Disaster relief 1.5 1.3 1.8 1.0 Wide Area Monitoring 0.8 1.5 0.7 0.9 In this embodiment, in order to enable the algorithm to adapt to the environment, it is necessary to first calculate the environmental complexity index, the core of which is to integrate four sub-indicators: obstacle distribution, meteorological interference, spatial constraints and electromagnetic environment.

[0089] In a preferred embodiment, calculating the environment complexity index further includes:

[0090] S221: Calculate the complexity of obstacle distribution The expression is as follows: In the formula, To determine the number of obstacles in the scanned area, For the sensor scan volume, For drones to the first The closest distance to each obstacle The standard deviation of the obstacle distance distribution. , , These are the weighting coefficients.

[0091] S222: Calculate the intensity of meteorological disturbances The expression is as follows: In the formula, For real-time wind speed, For the maximum wind resistance speed of the drone, For wind speed change rate, These are the weighting coefficients.

[0092] S223: Calculate the spatial constraint coefficients The expression is as follows: In the formula, To achieve the minimum width of the passable channel, , For the wingspan of the drone, This is a safety margin for height above the ground / top. , These are the weighting coefficients.

[0093] S224: Calculating Electromagnetic Environment Index The expression is as follows: In the formula, For communication interference signal strength, For wireless communication bit error rate, These are the weighting coefficients;

[0094] S225: Yes Perform normalization; the specific expression is:

[0095] S226: Substitute the normalized result into the environmental complexity index formula to complete the calculation.

[0096] S23: Introducing an identity matrix to maintain the positive definiteness of the Hessian matrix inverse, and using an improved BFGS formula to update the approximate value of the Hessian matrix inverse, the specific expression is as follows: In the formula, For the first The inverse approximation of the Hessian matrix in the next iteration , The gradient change vector, Update the vector for the parameters. It is the identity matrix. This represents the environmental complexity factor.

[0097] S24: Dynamically adjust the optimization step size based on the environmental complexity index. The specific expression is as follows: In the formula, This is the initial step size.

[0098] S25: Combine the adjusted step size with the updated Hessian matrix inverse approximation to complete trajectory optimization and output the optimal trajectory set.

[0099] In this embodiment, the inverse approximation of the Hessian matrix is ​​initialized. Set the initial step size The iteration terminates when the gradient norm is less than 10. −4 Or the number of iterations is ≥100. Each iteration is executed according to the following process: First, update the inverse approximation of the Hessian matrix according to the improved BFGS formula; second, dynamically adjust the optimization step size according to the environmental complexity index. The step size is dynamically adjusted as the environmental complexity increases, and the step size decreases to ensure the stability of the iteration process; finally, calculate the parameter update amount. The process involves verifying whether the new trajectory satisfies the UAV dynamics constraints; if not, the trajectory is trimmed and corrected. This iteration is repeated until the termination condition is met, ultimately outputting the optimal trajectory set that satisfies all constraints.

[0100] S3: Construct a Euclidean symbolic distance field using the standardized environment dataset, perform a collision risk assessment on the sampling points on the optimal trajectory set, and output the trajectory safety determination result.

[0101] Specifically, step S3 is executed as follows:

[0102] S31: Convert the obstacle point cloud data of the standardized environmental dataset to a resolution of [resolution value missing].

[0103] A 3D raster map using heuristic functions

[0104]

[0105] Generate the initial path, where, The actual cost from the starting point to node n. To heuristically estimate the cost from n to the target point, a smooth trajectory is generated with the objective of minimizing the time derivative of acceleration. The specific expression is as follows: In the formula, The function represents the three-dimensional spatial trajectory parameters of the UAV, where T represents the complete execution time, t represents the current time, and d represents the preset trajectory smoothing coefficient.

[0106] In this embodiment, obstacle point cloud data from a standardized environmental dataset is converted into a 3D raster map. The raster resolution is set to 0.1m × 0.1m × 0.1m, and the raster status is divided into three categories: occupied, idle, and unknown. A raster is marked as occupied when the number of obstacle point clouds within it is ≥3, as idle when there are no point clouds, and as unknown when the number of point clouds is <3. Then, based on the raster map, the following is applied... The algorithm generates the initial path, and its heuristic function is: ;in, This represents the actual path length from the starting point to node n. Let n be the Euclidean distance from node n to the destination. The heuristic factor is set to 1.0 to ensure the algorithm's optimality. The waypoint spacing of the initial path is set to 0.5m to balance path accuracy and computational efficiency.

[0107] To reduce actuator vibration and improve flight stability, The initial path is optimized using minimum jerk trajectory optimization. Specifically, a fifth-order polynomial fitting is used to achieve trajectory smoothing. The position, velocity, and acceleration constraints of waypoints (velocity and acceleration are continuous at waypoints) are substituted to solve for the polynomial coefficients, generating a smooth trajectory that satisfies the dynamic characteristics.

[0108] S32: Construct the Euclidean symbolic distance field for any point in space. The specific expression for calculating the distance value is as follows: In the formula, For the spatial location of obstacles, Characteristic points The positional relationship with obstacles.

[0109] S33: Perform ESDF value detection on each sampling point of the optimal trajectory set. If the ESDF value is less than the safety threshold... If an external collision risk is detected, the minimum distance between drones within the cluster will be checked; if it is less than the safe formation spacing... If an internal collision risk is detected, a trajectory safety assessment result will be output.

[0110] In this embodiment, a fast-moving method is used to construct the ESDF. The specific process is as follows: First, initialize the distance field, setting the distance values ​​of all occupied grids to 0, and setting the distance values ​​of idle and unknown grids to infinity; then, using the occupied grid as the source point, expand outwards in 6-neighborhoods, updating the distance values ​​of adjacent grids; finally, add a symbol to the distance values, with negative distance values ​​for occupied areas and positive distance values ​​for idle areas. In practical applications, the overall construction time of the ESDF is controlled within 5ms, which can meet the real-time requirements.

[0111] Based on this, the optimal trajectory set is sampled and detected at a frequency of 20Hz, with the sampling period consistent with the UAV control period: External collision detection: a safety threshold is set. =0.5m, when the ESDF value of the sampling point < When an external collision risk is identified, a ESDF value < 0 indicates a high risk, and a ESDF value ≤ 0 indicates a low risk. The risk level is medium; internal collision detection: set safe spacing for formation. =1.5m, calculate the Euclidean distance between any two drones in the cluster in real time, when the distance < At that time, it was determined that there was a risk of internal collision.

[0112] Therefore, the final output trajectory safety judgment result is: if there are no risk points, it is judged as safe; if there are medium risk points, local fine-tuning of the trajectory is triggered; and if there are high risk points, trajectory replanning is initiated.

[0113] S4: Based on the cluster state dataset and trajectory safety determination results, the formation topology is mathematically represented and eigenvalue modulated using the Laplace matrix to generate target formation parameters that adapt to environmental constraints.

[0114] Specifically, step S4 is executed as follows:

[0115] S41: Constructing an undirected topology graph based on cluster communication connection relationships ,in For a collection of drone nodes, This is the set of communication edges between nodes.

[0116] S42: Based on the topology diagram Calculate the degree matrix Adjacency Matrix This leads to the Laplace matrix, which is expressed as follows:

[0117]

[0118] In this embodiment, an undirected topological graph G=(V,E) is constructed based on the cluster communication topology, where the node set V represents the numbers of all UAVs, and the rule for determining the edge set E is: when the communication distance between two UAVs is ≤50m, a communication edge is established. The degree matrix D and adjacency matrix A are calculated based on the topological graph: the degree matrix D is a diagonal matrix, and the diagonal element Dii represents the number of communication edges for node i; in the adjacency matrix A, if (i,j)∈E, then Aij=1, otherwise Aij=0. The Laplace matrix is ​​calculated using the formula L=D−A. S43: Eigenvalue decomposition is performed on the Laplace matrix, the specific expression is:

[0119]

[0120] In the formula, It is an eigenvalue diagonal matrix. This is the eigenvector matrix.

[0121] S44: Utilize The eigenvectors corresponding to non-zero eigenvalues ​​represent the formation topology, while the eigenvectors corresponding to zero eigenvalues ​​represent the overall translational characteristics of the formation, generating initial formation encoding data.

[0122] In this embodiment, the QR algorithm is used to perform eigenvalue decomposition on the Laplacian matrix, and the formation encoding is realized by using eigenvalues ​​and eigenvectors: the eigenvector q0 corresponding to the zero eigenvalue λ0=0 represents the overall translational characteristics of the formation; the eigenvectors corresponding to the non-zero eigenvalues ​​λ1,...,λm−1 represent the topological structure of the formation, such as λ1 reflecting the compactness of the formation and λ2 reflecting the symmetry of the formation. The mathematical representation of any formation can be realized based on the linear combination of eigenvectors.

[0123] S45: Based on the environmental constraints in the trajectory safety determination result, perform proportional scaling on non-zero feature values ​​to finally generate the target formation parameters. The specific expression is as follows:

[0124]

[0125] In the formula, These are the original eigenvalues. Adjust the scaling factor to fit the environment.

[0126] In this embodiment, the scaling factor `scale_factor` is determined based on the environmental constraints in the trajectory safety determination result: when the environmental channel width `Wpath` is greater than 1.5 times the current formation width, `scale_factor` = 1.0, maintaining the original formation; when `Wpath` is 1.0 to 1.5 times the current formation width, `scale_factor` = `Wpath` / (1.2 × current formation width), scaling the formation proportionally; when `Wpath` is less than 1.0 times the current formation width, `scale_factor` = `Wpath` / (1.0 × current formation width), and formation topology reconstruction (converting to a single-column formation) is triggered. After this, the non-zero eigenvalues ​​of the Laplacian matrix are modulated, and a new adjacency matrix is ​​inferred based on the modulated eigenvalues ​​to generate target formation parameters adapted to environmental constraints, including the relative spacing, arrangement angle, and waypoint offset of each UAV.

[0127] S5: The optimal trajectory set and target formation parameters are converted into low-level flight control commands to drive the UAV to perform flight actions, while collecting real-time motion status data.

[0128] Specifically, step S5 is executed as follows:

[0129] S51: Convert the optimal trajectory set and target formation parameters into attitude control commands and power drive commands for each UAV.

[0130] First, the optimal trajectory set is converted into attitude control commands. The desired roll, pitch, and yaw angles of the UAV are obtained through inverse kinematics calculations. The roll and pitch angles are determined by the acceleration components along the trajectory tangent, while the yaw angle is determined by formation direction constraints. Next, the target formation parameters are converted into power drive commands. Based on the UAV's position deviation (the difference between its actual position and the desired formation position), a PID controller calculates the motor speed increment. Finally, the attitude and power commands are integrated into a 50Hz PWM signal, which serves as the control input for the underlying actuators. Control commands are distributed through a cluster communication network, employing a command-acknowledgment mechanism to ensure reliability. The receiving end must return an acknowledgment signal within 10ms; if not received, a retransmission is triggered. The number of retransmissions is ≤3. If the retransmission still fails, a backup communication link is switched.

[0131] S52: Transmits control commands to the UAV actuators, driving the rotor system to complete the corresponding flight maneuvers.

[0132] After receiving PWM commands, the UAV's onboard actuator module drives the electronic speed controller to adjust the motor speed, thereby controlling the rotor thrust and torque. When increased lift is needed, the ESC increases the motor duty cycle; when attitude adjustment is needed, differential adjustment of each rotor speed achieves roll, pitch, or yaw. The actuator has built-in overcurrent and overtemperature protection. When the motor current exceeds 120% of the rated value or the temperature exceeds 80°C, it automatically reduces power and reports a fault to ensure hardware safety.

[0133] S53: Collect the pose and velocity of the UAV after executing its actions through the state acquisition device, and update the cluster state dataset after preprocessing.

[0134] In this embodiment, the UAV collects state data after the execution of actions through onboard sensors: the IMU collects motion attitude data every 20ms, the GPS collects position data every 100ms, and the motor sensors collect speed and current data every 50ms. After the raw data undergoes denoising, registration, and normalization in the preprocessing process of step S1, it is updated to the cluster state dataset at a period of 20ms to provide the latest state input for the next round of decision-making.

[0135] In a preferred embodiment, such as Figure 2-3 As shown, this invention also proposes a UAV swarm formation control system based on a multi-level intelligent algorithm. The system includes: a perception layer with an environmental perception module, a state estimation module, and a communication module; a decision layer with a trajectory optimization module, a collision detection module, and a formation representation module; and an execution layer with a flight control module and an actuator module. The perception layer, the decision layer, and the execution layer are configured to jointly execute the UAV swarm formation control method based on the multi-level intelligent algorithm described above.

[0136] Other embodiments or specific implementations of the UAV swarm formation control system based on multi-level intelligent algorithms of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0137] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0138] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0139] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for controlling unmanned aerial vehicle (UAV) swarm formations based on a multi-level intelligent algorithm, characterized in that, The method includes the following steps: The system acquires external environmental monitoring data and motion status data of the UAV swarm, performs multimodal fusion preprocessing on the environmental monitoring data and motion status data, and achieves global sharing of the dataset through the swarm communication network to obtain a synchronized standardized environmental dataset and swarm status dataset. Based on the standardized environment dataset and cluster state dataset, a multi-objective comprehensive cost function is constructed, and an improved quasi-Newton method is used to optimize the initial trajectory of the cluster formation to obtain the optimal trajectory set that satisfies the dynamic constraints. Specifically, an improved quasi-Newton method is used to optimize the initial trajectory of the swarm formation to obtain the optimal trajectory set that satisfies the dynamic constraints, including: Based on the standardized environment dataset, an environment complexity index is calculated, and then an environment complexity factor is derived from the environment complexity index. The specific expression is as follows: ； In the formula, As the basic weight for environmental indicators, For obstacle distribution complexity, For meteorological interference intensity, For spatial constraint coefficients, Electromagnetic environment index; The basic regularization coefficient, Let the current gradient norm be... As an environmental complexity factor, This is the sensitivity coefficient; An identity matrix is ​​introduced to maintain the positive definiteness of the Hessian matrix inverse. An improved BFGS formula is used to update the approximate value of the Hessian matrix inverse. The specific expression is as follows: ； In the formula, For the first The inverse approximation of the Hessian matrix in the next iteration , The gradient change vector, Update the vector for the parameters. It is the identity matrix. Environmental complexity factor; The optimization step size is dynamically adjusted based on the environmental complexity index, as expressed in the following expression: In the formula, This is the initial step size; The trajectory optimization is completed by combining the adjusted step size with the updated inverse approximation of the Hessian matrix, and the optimal trajectory set is output. The computational environment complexity metrics include: Computational obstacle distribution complexity The expression is as follows: ； In the formula, To determine the number of obstacles in the scanned area, For the sensor scan volume, For drones to the first The closest distance to each obstacle The standard deviation of the obstacle distance distribution. , , These are the weighting coefficients; Calculate the intensity of meteorological disturbance The expression is as follows: ； In the formula, For real-time wind speed, For the maximum wind resistance speed of the drone, For wind speed change rate, These are the weighting coefficients; Calculate spatial constraint coefficients The expression is as follows: ； In the formula, To achieve the minimum width of the passable channel, , For the wingspan of the drone, This is a safety margin for height above the ground / top. , These are the weighting coefficients; Calculate the electromagnetic environment index The expression is as follows: ； In the formula, For communication interference signal strength, For wireless communication bit error rate, These are the weighting coefficients; right Perform normalization; the specific expression is: ； Substitute the normalized result into the environmental complexity index formula to complete the calculation; Using the standardized environmental dataset, a Euclidean symbolic distance field is constructed, and a collision risk assessment is performed on the sampling points on the optimal trajectory set to output the trajectory safety determination result. Based on the cluster state dataset and trajectory safety determination results, mathematical representation and eigenvalue modulation of the formation topology are performed using the Laplace matrix to generate target formation parameters that adapt to environmental constraints. The optimal trajectory set and target formation parameters are converted into low-level flight control commands to drive the UAV to perform flight maneuvers, while simultaneously collecting real-time motion status data.

2. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, The external environment monitoring data includes obstacle distribution point cloud, environmental channel width, real-time wind speed, and communication interference intensity. The motion state data includes the UAV's pose, speed, and acceleration.

3. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, Multimodal fusion preprocessing is performed on the environmental monitoring data and motion state data. Global sharing of the datasets is achieved through a cluster communication network to obtain synchronized, standardized environmental datasets and cluster state datasets. Specifically, this includes: The raw environmental monitoring data and raw motion state data are preprocessed by denoising, registration and normalization to obtain standardized environmental datasets and cluster state datasets. The standardized environmental dataset and cluster status dataset are transmitted to the decision-making units of all UAVs in the cluster using the established cluster communication network, ensuring data synchronization.

4. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, Based on the standardized environment dataset and cluster state dataset, a multi-objective comprehensive cost function is constructed, specifically including: Constructing a path length cost Energy consumption cost Security costs and formation stability cost The multi-objective comprehensive cost function is expressed as follows: In the formula, The preset weighting coefficients are adjusted for adaptability based on different task types.

5. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, Using the standardized environmental dataset, a Euclidean symbolic distance field is constructed. Collision risk assessment is performed on the sampled points on the optimal trajectory set, and the trajectory safety determination result is output, specifically including: The obstacle point cloud data of the standardized environmental dataset is converted to a resolution of [resolution value missing]. A 3D raster map using heuristic functions Generate the initial path, where, The actual cost from the starting point to node n. To heuristically estimate the cost from n to the target point, a smooth trajectory is generated with the objective of minimizing the time derivative of acceleration. The specific expression is as follows: In the formula, The function represents the three-dimensional spatial trajectory parameters of the UAV, where T represents the complete execution time, t represents the current time, and d represents the preset trajectory smoothing coefficient. Construct the Euclidean symbolic distance field for any point in space. The specific expression for calculating the distance value is as follows: In the formula, For the spatial location of obstacles, Characteristic points Positional relationship with obstacles; Perform ESDF value detection on each sampling point of the optimal trajectory set. If the ESDF value is less than the safety threshold... If an external collision risk is detected, the minimum distance between drones within the cluster will be checked; if it is less than the safe formation spacing... If an internal collision risk is detected, a trajectory safety assessment result will be output.

6. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, Based on the cluster state dataset and trajectory safety determination results, the formation topology is mathematically represented and its eigenvalues ​​modulated using the Laplace matrix to generate target formation parameters adapted to environmental constraints. Specifically, this includes: Constructing an undirected topology graph based on cluster communication connection relationships ,in For a collection of drone nodes, This is the set of communication edges between nodes; According to the topology diagram Calculate the degree matrix Adjacency Matrix This leads to the Laplace matrix, which is expressed as follows: The eigenvalue decomposition of the Laplacian matrix is ​​expressed as follows: In the formula, It is an eigenvalue diagonal matrix. The eigenvector matrix; use The eigenvectors corresponding to non-zero eigenvalues ​​represent the formation topology, and the eigenvectors corresponding to zero eigenvalues ​​represent the overall translational characteristics of the formation, generating initial formation coding data. Based on the environmental constraints in the trajectory safety determination result, non-zero feature values ​​are scaled proportionally to generate the target formation parameters, the specific expression of which is: In the formula, These are the original eigenvalues. Adjust the scaling factor to fit the environment.

7. The UAV swarm formation control method based on multi-level intelligent algorithms as described in claim 1, characterized in that, The optimal trajectory set and target formation parameters are converted into low-level flight control commands to drive the UAV to perform flight maneuvers, while simultaneously collecting real-time motion state data, specifically including: The optimal trajectory set and target formation parameters are converted into attitude control commands and power drive commands for each UAV. The control commands are transmitted to the drone's actuators, which drive the rotor system to complete the corresponding flight maneuvers. The pose and velocity of the drone after its actions are collected by the state acquisition device, and the cluster state dataset is updated after preprocessing.

8. A drone swarm formation control system based on a multi-level intelligent algorithm, characterized in that, The system includes: a perception layer having an environment perception module, a state estimation module, and a communication module; a decision layer having a trajectory optimization module, a collision detection module, and a formation representation module; and an execution layer having a flight control module and an actuator module; wherein the perception layer, the decision layer, and the execution layer are configured to jointly execute the UAV swarm formation control method based on a multi-level intelligent algorithm as described in any one of claims 1-7.

Images