Intelligent Cooperative Control Method and System Based on Unmanned Aerial Vehicle Swarm

Abstract

This invention discloses an intelligent cooperative control method and system based on UAV swarms, belonging to the field of intelligent control technology for UAV swarms. The method includes collecting multi-source data to construct a sparse tensor, defining UAVs as nodes, and constructing a dynamic category graph. The nodes of the dynamic category graph are mapped to manifold points of a Riemannian manifold, UAV coordinates are converted to grid indices, an objective function is defined, the Riemann gradient of the objective function is calculated, and the manifold points are updated using exponential mapping to obtain updated spatial coordinates and task vectors. A weighted linear combination is used to generate control command vectors, which are then converted into control commands and executed. By constructing a sparse tensor and a dynamic category graph, and combining Riemannian manifold optimization to balance task priority and cooperative constraints, task coverage and cooperative robustness are improved. The combination of Riemannian gradient optimization and exponential mapping updates enhances autonomous cooperative capabilities.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for unmanned aerial vehicle (UAV) swarms, and in particular to intelligent collaborative control methods and systems based on UAV swarms. Background Technology

[0002] With advancements in sensor technology, embedded computing capabilities, and communication networks, unmanned aerial vehicle (UAV) swarms are capable of performing complex collaborative tasks such as search and rescue, environmental monitoring, and precision agricultural management. Information fusion based on multi-source sensors (such as vision, radar, and inertial measurement units), collaborative path planning based on distributed algorithms, and behavior control methods based on swarm intelligence have become the mainstream technical approaches for UAV swarm collaborative control. However, existing collaborative control methods still face many challenges in the face of highly dynamic, nonlinear, and complex and ever-changing mission environments.

[0003] Existing UAV swarm cooperative control technology still has several shortcomings. The existing technology is based on Euclidean geometry navigation method, which is difficult to achieve effective task allocation and path planning under the drive of high-dimensional information flow, and the system flexibility is low. Existing optimization methods are mostly based on Euclidean space, ignoring the nonlinear geometric characteristics of the task space, and cannot effectively balance task priority and cooperative constraints, thus affecting the global optimality of path planning. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent cooperative control method and system based on UAV swarms, which solves the problems of existing navigation methods based on Euclidean geometry, which are difficult to achieve effective task allocation and path planning under high-dimensional information flow, and have low system flexibility. Existing optimization methods are mostly based on Euclidean space, ignoring the nonlinear geometric characteristics of the task space, and cannot effectively balance task priority and cooperative constraints, thus affecting the global optimality of path planning.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms, comprising,

[0008] Collect and preprocess multi-source data to construct a sparse tensor. Use the finite difference method to calculate the partial derivatives of each dimension in the sparse tensor. Use the vector synthesis method to combine the partial derivatives as the gradient vector field of information entropy. Define the UAV as a node and construct a dynamic category graph.

[0009] The nodes of the dynamic category graph are mapped to the popular points of the Riemannian manifold, the UAV coordinates are converted into grid indices, the mission intensity impact of the UAV is calculated, the metric tensor is calculated, positive definite regularization is added to the metric tensor, the objective function is defined, the Riemann gradient of the objective function is calculated, and the manifold points are updated using exponential mapping to obtain the updated spatial coordinates and mission vector.

[0010] A weighted linear combination is used to generate a control command vector, which is then converted into control commands and executed.

[0011] Build a visual interface to display control commands.

[0012] As a preferred embodiment of the intelligent cooperative control method based on UAV swarms described in this invention, the step of collecting multi-source data, constructing a sparse tensor, defining UAVs as nodes, and constructing a dynamic category graph includes:

[0013] Use smart sensors to collect multi-source data from drone swarms and perform noise reduction and normalization processing.

[0014] The multi-source data includes images, 3D point clouds, spatial coordinates, velocity, and acceleration data;

[0015] Extended Kalman filtering is used to fuse spatial coordinates, velocity, and acceleration data. K-means clustering is used to cluster the fused data. A pre-trained Swin-Transformer model is used to recognize the image and output the corresponding state variables. Spatial coordinates, 3D point cloud, velocity, acceleration, clustering results, and state labels are stacked using tensor stacking to construct a sparse tensor. The non-zero elements of the sparse tensor are summed to calculate the task intensity probability. The information entropy is calculated based on the task intensity probability. The partial derivatives of each dimension in the sparse tensor are calculated using the finite difference method based on the information entropy. The partial derivatives are combined using vector synthesis to form the gradient vector field of the information entropy.

[0016] Extract the target grid coordinates from the sparse tensor, convert the grid coordinates into actual geographic coordinates, calculate the direction vector from the UAV to the target location, normalize the direction vector, and obtain the task vector.

[0017] Calculate the control mapping function between UAVs, set the filtering threshold using the fixed threshold method, and define the filtered control mapping function as a control state mapping function.

[0018] The drones are defined as nodes, and control states with a value greater than 0 are defined as edges. An initial category graph is constructed. A change threshold is set using statistical analysis. The Euclidean distance between the spatial position of each drone and the previous time is calculated. Euclidean distances greater than the change threshold are filtered out and marked as affected control states. The control states are then recalculated. Otherwise, the control states from the previous time are retained, and a dynamic category graph is generated.

[0019] As a preferred embodiment of the intelligent cooperative control method based on UAV swarms described in this invention, the steps of mapping nodes of the dynamic category graph to popular points of the Riemannian manifold, converting UAV coordinates to grid indices, and defining the objective function include:

[0020] Based on drones, a Riemannian manifold is constructed. The Euclidean distance between points of the manifold is calculated using the Euclidean distance formula and defined as the Euclidean metric. The initial metric tensor is set using the Euclidean metric, and the nodes of the dynamic category graph are mapped to the manifold points of the Riemannian manifold.

[0021] We construct the adjacency matrix on the manifold by using the edges of the dynamic category graph as adjacency constraints.

[0022] Using a grid mapping table, the drone coordinates are converted into grid indices, the mission intensity impact of the drone is calculated, the metric tensor is calculated, and positive definite regularization is applied to the metric tensor.

[0023] We construct a weighted objective function and define the objective function.

[0024] As a preferred embodiment of the intelligent cooperative control method based on UAV swarms described in this invention, the calculation of the Riemann gradient of the objective function, updating the manifold points using an exponential mapping, and obtaining the updated spatial coordinates and task vector includes:

[0025] The gradient vector and metric tensor are combined to compute the Riemann gradient of the objective function, and the manifold points are updated using an exponential mapping based on the Riemann gradient.

[0026] Use vector decomposition to decompose the updated spatial coordinates and task vectors from the updated manifold points.

[0027] As a preferred embodiment of the intelligent cooperative control method based on UAV swarms described in this invention, the step of generating control command vectors using weighted linear combination includes:

[0028] The position increment is calculated using vector subtraction, and the control command vector is generated using a weighted linear combination and then normalized.

[0029] As a preferred embodiment of the intelligent cooperative control method based on UAV swarms described in this invention, the step of converting the control command vector into control commands and executing them includes:

[0030] The normalized control command vector is converted into control commands using a vector set encapsulation method.

[0031] Extract the adjacency matrix of the dynamic category graph, allocate communication channels for the UAV using a wireless communication protocol, send control commands to the corresponding UAV, and the UAV receives and executes the control commands.

[0032] As a preferred embodiment of the intelligent collaborative control method based on UAV swarms described in this invention, the step of constructing a visual interface to display control commands includes:

[0033] The front-end framework React.js is used to build a visual interface to visualize the control commands obtained from the analysis and the multi-source data collected.

[0034] Users who have completed real-name verification are allowed to view this information.

[0035] Secondly, the present invention provides an intelligent collaborative control system based on unmanned aerial vehicle (UAV) swarms, comprising,

[0036] The graph collection module is used to collect and preprocess multi-source data, construct a sparse tensor, calculate the partial derivative of each dimension of the sparse tensor using the finite difference method, combine the partial derivatives using the vector synthesis method as the gradient vector field of information entropy, define the UAV as a node, and construct a dynamic category graph.

[0037] The target update module is used to map the nodes of the dynamic category graph to the popular points of the Riemannian manifold, convert the UAV coordinates into grid indices, calculate the mission intensity impact of the UAV, calculate the metric tensor, add a regularization term to the metric tensor with positive definite regularization, define the objective function, calculate the Riemann gradient of the objective function, update the manifold points using exponential mapping, and obtain the updated spatial coordinates and mission vector.

[0038] The instruction execution module is used to generate control instruction vectors using weighted linear combinations, convert the control instruction vectors into control instructions, and execute them.

[0039] The visualization module is used to build a visual interface to display control commands.

[0040] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent cooperative control method based on unmanned aerial vehicle swarms as described in the first aspect of the present invention.

[0041] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent cooperative control method based on unmanned aerial vehicle swarms as described in the first aspect of the present invention.

[0042] The beneficial effects of this invention are as follows: By collecting and preprocessing multi-source data, a sparse tensor is constructed. The partial derivatives of each dimension in the sparse tensor are calculated using the finite difference method. The partial derivatives are combined using vector synthesis to form the gradient vector field of information entropy. UAVs are defined as nodes, and a dynamic category graph is constructed. The nodes of the dynamic category graph are mapped to the manifold points of the Riemannian manifold. The UAV coordinates are converted into grid indices, the task intensity impact of the UAV is calculated, the metric tensor is calculated, a positive definite regularization term is added to the metric tensor, an objective function is defined, the Riemann gradient of the objective function is calculated, and the manifold points are updated using exponential mapping to obtain the updated spatial coordinates and task vector. A weighted linear combination is used to generate a control command vector, which is then converted into control commands and executed. This invention solves the shortcomings of traditional methods in complex dynamic environments, significantly improves task coverage and collaborative robustness, and enhances autonomous collaborative capabilities. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of the intelligent collaborative control method based on UAV swarm in Example 1.

[0045] Figure 2 This is a schematic diagram of the intelligent collaborative control system based on unmanned aerial vehicle (UAV) swarm in Example 1.

[0046] Figure 3 This is a schematic diagram of the collaborative control in the intelligent collaborative control method based on UAV swarm in Example 1. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Example 1, referring to Figures 1 to 3 This is the first embodiment of the present invention, which provides an intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms, including the following steps:

[0051] S1. Collect and preprocess multi-source data, construct a sparse tensor, use the finite difference method to calculate the partial derivatives of each dimension in the sparse tensor, use the vector synthesis method to combine the partial derivatives as the gradient vector field of information entropy, define the UAV as a node, and construct a dynamic category graph.

[0052] Specifically, multi-source data is collected and preprocessed to construct a sparse tensor. The partial derivatives of each dimension in the sparse tensor are calculated using the finite difference method. Vector synthesis is used to combine the partial derivatives as the gradient vector field of the information entropy. The UAV is defined as a node, and a dynamic category graph is constructed, including:

[0053] Use smart sensors to collect multi-source data from drone swarms and perform noise reduction and normalization processing.

[0054] The multi-source data includes images, 3D point clouds, spatial coordinates, velocity, and acceleration data;

[0055] The intelligent sensors include cameras, lidar, GPS, and IMU (inertial measurement unit) sensors;

[0056] Extended Kalman filtering is used to fuse spatial coordinates, velocity, and acceleration data. K-means clustering is used to cluster the fused data (the number of clusters is set using the silhouette coefficient method). A pre-trained Swin-Transformer model is used to recognize the image and output the corresponding state variables. Spatial coordinates, 3D point cloud, velocity, acceleration, clustering results, and state labels are stacked using tensor stacking to construct a sparse tensor. The non-zero elements of the sparse tensor are summed to calculate the task intensity probability. The information entropy is calculated based on the task intensity probability. Based on the information entropy, the partial derivatives of each dimension in the sparse tensor are calculated using the finite difference method. The partial derivatives are combined using vector synthesis to form the gradient vector field of the information entropy.

[0057] Extract the target grid coordinates from the sparse tensor, corresponding to the largest intensity cell in the tensor, convert the grid coordinates to actual geographic coordinates (through a grid mapping table), calculate the direction vector from the UAV to the target location, normalize the direction vector, and obtain the mission vector.

[0058] Calculate the control mapping function between UAVs, set the filtering threshold using the fixed threshold method, and define the filtered control mapping function as a control state mapping function.

[0059] Define the drone as a node and the control morphism greater than 0 as an edge. Construct an initial category graph. Use statistical analysis to set a change threshold. Calculate the Euclidean distance between the spatial position of each drone and the previous time. Filter out Euclidean distances greater than the change threshold and mark them as affected control morphisms. Recalculate the control morphisms. Otherwise, retain the control morphisms from the previous time and generate a dynamic category graph.

[0060] The task strength probability is calculated using a normalized probability distribution based on the sparse task tensor, with the following formula:

[0061] ,

[0062] in Let be the task strength probability of the sparse tensor at time t. For sparse tensors, Let be the sum of intensities, represent the sum of non-zero elements of the sparse tensor, i is the index of the task type dimension, j is the index of the spatial coordinates, and k is the index of the state variables.

[0063] The information entropy of sparse tensors is calculated using Shannon entropy based on task intensity probability, with the following formula:

[0064] ,

[0065] in Let be the information entropy of the sparse tensor at time t;

[0066] The formula for extracting target grid coordinates is:

[0067] ,

[0068] in Let t be the target grid coordinates of the UAV;

[0069] The formula for calculating the direction vector from the UAV to the target position is:

[0070] ,

[0071] in Let l be the direction vector of the l-th UAV. Let t be the actual geographic coordinates of the UAV. Let J be the spatial coordinates of the Jth UAV;

[0072] The direction vector is normalized using vector normalization to obtain the task vector. If the task vector is 0, it means the UAV has reached the target. Let... The formula is:

[0073] , ,

[0074] in Let l be the task vector of the l-th UAV;

[0075] The control mapping function between UAVs is calculated using the following formula:

[0076] , ,

[0077] in Let be the control mapping function, representing the l-th UAV. up to the uth drone The strength of control dependency The Euclidean distance between the l-th and u-th drones is calculated using the Euclidean distance formula. It is a very small constant. Let u be the task vector of the u-th drone. Let time t be the l-th drone up to the uth drone The control morphism represents the value of the control dependency. This is the filtering threshold.

[0078] This invention integrates multi-source heterogeneous data into sparse tensors, combining information entropy and gradient vector fields to guide task allocation for UAV swarms. Existing technologies typically process data separately (such as images or point clouds) or use high-dimensional dense tensors, resulting in high computational complexity. This invention efficiently processes multi-source data, generates task-driven directions, and is suitable for dynamic task allocation. It reduces storage requirements through sparse tensors, quantifies task uncertainty through information entropy, provides global driving directions through gradient vector fields, and uses a dynamic category graph to update control morphisms in real time to adapt to topological changes in the UAV swarm. Vector synthesis ensures the coordinated expression of gradient information in each dimension, avoiding path oscillations or execution ambiguities caused by single-dimensional optimization. The Transformer, as one of the best-performing structures in current image recognition, enhances the contextual modeling capability of information, making it particularly suitable for identifying target types under complex conditions such as low light and occlusion. The control morphism uses a fixed threshold method to filter effective edges, avoiding control oscillations caused by overly dense graph structures, while retaining the main control paths. The dynamic update mechanism ensures that the structure responds to task evolution or positional disturbances, thereby achieving self-organized updates of control relationships within the region and ensuring the coordinated stability of large-scale UAV swarms under dynamic task-driven conditions.

[0079] S2. Map the nodes of the dynamic category graph to the popular points of the Riemannian manifold, convert the UAV coordinates to grid indices, calculate the mission intensity impact of the UAV, calculate the metric tensor, add a regularization term to the metric tensor with positive definite regularization, define the objective function, calculate the Riemann gradient of the objective function, update the manifold points using exponential mapping, and obtain the updated spatial coordinates and mission vector.

[0080] Specifically, the nodes of the dynamic category graph are mapped to the popular points of the Riemannian manifold, the UAV coordinates are converted to grid indices, the mission intensity impact of the UAV is calculated, the metric tensor is calculated, positive definite regularization is added to the metric tensor, and the objective function is defined, including:

[0081] Based on drones, the manifold dimension is defined by the following formula:

[0082] ,

[0083] in 3 represents the dimension of the manifold, and 3 represents the dimension of the spatial coordinates;

[0084] The formula for constructing popular points is:

[0085] ,

[0086] in For the l-th popular point at time t, Let be the spatial coordinates of the l-th UAV at time t. For local coordinate space;

[0087] The Euclidean distance between points on a manifold is calculated using the Euclidean distance formula and defined as the Euclidean metric. The initial metric tensor is then set using the Euclidean metric.

[0088] Mapping nodes of a dynamic category graph to flow points of a Riemannian manifold is done using the following formula:

[0089] ,

[0090] in Let f(l) be the popular point at time t, and let f(l) be the popular point on the manifold corresponding to the l-th UAV, containing spatial coordinates and mission vector. It is a Riemannian manifold;

[0091] We construct the adjacency matrix on the manifold by using the edges of the dynamic category graph as adjacency constraints.

[0092] Using a grid mapping table, the UAV coordinates are converted into grid indices, and the mission intensity impact of the UAV is calculated using the following formula:

[0093] ,

[0094] in To account for the impact of the task intensity of the l-th UAV at time t, the index function is 1 when only the grid index matching position of the j-th UAV is accumulated;

[0095] The formula for calculating the metric tensor is:

[0096] ,

[0097] in For popular points The metric tensor components at point Q and R are the row and column indices of the manifold dimension, respectively. Euclidean measurement, Let be the information entropy of the sparse tensor at time t. At the location of the drone The gradient vector (calculated using the finite difference method);

[0098] Positive definite regularization adds a regularization term to the metric tensor, with the following formula:

[0099] ,

[0100] Where I is the identity matrix;

[0101] Using a weighted objective function, the objective function is defined as follows:

[0102] ,

[0103] in Let be the objective function, representing the total cost of cooperative optimization of the drone swarm. Let be the geodesic distance between the popular points at position l and position u on the manifold (calculated using Dijkstra's algorithm based on the metric).

[0104] By embedding drones and their relationships into a manifold, the system can naturally incorporate curvature and local geometric properties, making collaborative decision-making more robust in complex task environments. It is not limited by traditional Euclidean modeling and solves the problems of static modeling and inability to capture group dynamics in existing technologies. It improves the adaptability to dynamic tasks and uncertainties. Through matrix-based constraint expression, it is easy to introduce linear algebra and optimization tools, making the model more computationally efficient in numerical solutions. It overcomes the shortcomings of existing systems in handling dynamic relationships and lacking a unified optimization interface. The computational task intensity influence solves the problems of large computational volume and insufficient real-time performance in continuous optimization in existing technologies. At the same time, the task intensity influence significantly improves the accuracy of task allocation. The introduction of positive definiteness constraints for the metric tensor solves the problems of singular matrices and numerical divergence in high-dimensional optimization in existing technologies, ensuring the convergence of the optimization path.

[0105] Furthermore, the Riemann gradient of the objective function is calculated, and the manifold points are updated using the exponential mapping to obtain the updated spatial coordinates and task vector, including:

[0106] The Riemann gradient of the objective function is calculated by combining the gradient vector and the metric tensor, as shown in the formula:

[0107] ,

[0108] in For the objective function J at the popular point Riemann gradient at that point Let u be the popular point at time t;

[0109] The manifold points are updated using an exponential mapping based on the Riemann gradient, as follows:

[0110] ,

[0111] in For the updated manifold points, Step size (set using the line search method);

[0112] Vector decomposition is used to decompose the updated spatial coordinates and task vectors from the updated manifold points, as shown in the formula:

[0113] ,

[0114] in and These are the updated coordinate space and the task vector, respectively.

[0115] Defining gradients on the manifold ensures geometric consistency in the optimization steps, avoiding invalid or even destructive update directions. This solves the problem in existing technologies where the optimization direction does not fit the surface space, leading to convergence failure. Exponential mapping ensures that the update point remains on the manifold, eliminating the need for forced projection back to the manifold. This makes the optimization process more natural and efficient, avoiding the high-cost operation of frequent constraint projection required in existing Euclidean optimization. It improves computational efficiency and accuracy, ensuring a seamless transition from the abstract optimization space to the concrete execution space. This solves the problem in existing technologies where optimization results are difficult to interpret and directly apply to control instructions, thus enhancing the system's practicality.

[0116] S3. Use weighted linear combination to generate control command vectors, convert control command vectors into control commands and execute them;

[0117] Specifically, the control command vector is generated using a weighted linear combination, including:

[0118] The position increment (current UAV position minus updated coordinate space) is calculated using vector subtraction. A weighted linear combination is then used to generate the control command vector, which is subsequently normalized. The formula is as follows:

[0119] ,

[0120] in Let be the command vector of the UAV at time t, Assign weights to the task vectors (using rules of thumb).

[0121] like When only the position increment is used, if Using only task vectors ( ).

[0122] Vector subtraction provides a simple, low-error incremental acquisition method, which helps improve real-time performance and accuracy. Existing normalization techniques often employ single-objective optimization, making it difficult to simultaneously meet the needs of multiple tasks. Weighted linear combination provides a natural multi-task fusion mechanism, while normalization solves the problem of imbalanced control command scale, ensuring the overall stability of multi-UAV systems. Traditional multi-UAV communication schemes are usually based on static network topology, making it difficult to cope with the communication reconstruction requirements under dynamic tasks.

[0123] Furthermore, the control instruction vector is converted into control instructions and executed, including:

[0124] The normalized control command vector is converted into control commands using a vector set encapsulation method.

[0125] Extract the adjacency matrix of the dynamic category graph, allocate communication channels for the UAV using a wireless communication protocol, send control commands to the corresponding UAV, and the UAV receives and executes the control commands.

[0126] This solution addresses common issues in group communication, such as conflicts, delays, and packet loss, through dynamic adjustment of the adjacency matrix and protocol allocation, significantly improving communication efficiency and robustness.

[0127] S4. Construct a visual interface to display control commands;

[0128] Specifically, a visual interface is built to display control commands, including:

[0129] The front-end framework React.js is used to build a visual interface to visualize the control commands obtained from the analysis and the multi-source data collected.

[0130] Users who have completed real-name verification are allowed to view this information.

[0131] Data visualization often uses static pages or reloaded rendering, resulting in slow response and poor interactivity. This invention uses React.js to enable changes in instructions and multi-source data to be presented in real time with low latency on the front end, greatly improving operational efficiency and user experience. In existing drone swarm systems, most use open interfaces or access control based on simple passwords, which poses significant security risks. Real-name verification not only improves the system's security level but also provides a mechanism to support task tracing and responsibility allocation.

[0132] This embodiment also provides an intelligent collaborative control system based on drone swarms, including:

[0133] The graph collection module is used to collect and preprocess multi-source data, construct a sparse tensor, calculate the partial derivative of each dimension of the sparse tensor using the finite difference method, combine the partial derivatives using the vector synthesis method as the gradient vector field of information entropy, define the UAV as a node, and construct a dynamic category graph.

[0134] The target update module is used to map the nodes of the dynamic category graph to the popular points of the Riemannian manifold, convert the UAV coordinates into grid indices, calculate the mission intensity impact of the UAV, calculate the metric tensor, add a regularization term to the metric tensor with positive definite regularization, define the objective function, calculate the Riemann gradient of the objective function, update the manifold points using exponential mapping, and obtain the updated spatial coordinates and mission vector.

[0135] The instruction execution module is used to generate control instruction vectors using weighted linear combinations, convert the control instruction vectors into control instructions, and execute them.

[0136] The visualization module is used to build a visual interface to display control commands.

[0137] This embodiment also provides a computer device applicable to the intelligent cooperative control method based on UAV swarms, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent cooperative control method based on UAV swarms as proposed in the above embodiment.

[0138] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0139] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent cooperative control method based on UAV swarms as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0140] In summary, this invention collects and preprocesses multi-source data to construct a sparse tensor. It then uses the finite difference method to calculate the partial derivatives of each dimension of the sparse tensor and combines these partial derivatives using vector synthesis to form the gradient vector field of the information entropy. UAVs are defined as nodes, and a dynamic category graph is constructed. The nodes of the dynamic category graph are mapped to the manifold points of the Riemannian manifold. The UAV coordinates are converted to grid indices, and the task intensity impact of the UAV is calculated. A metric tensor is calculated, and positive definite regularization is applied to the metric tensor. An objective function is defined, and its Riemannian gradient is calculated. The manifold points are updated using exponential mapping to obtain the updated spatial coordinates and task vector. A weighted linear combination is used to generate control command vectors, which are then converted into control commands and executed. This invention addresses the shortcomings of traditional methods in complex dynamic environments, significantly improves task coverage and collaborative robustness, and enhances autonomous collaborative capabilities.

[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms, characterized in that: include, Collect and preprocess multi-source data to construct a sparse tensor. Use the finite difference method to calculate the partial derivatives of each dimension in the sparse tensor. Use the vector synthesis method to combine the partial derivatives as the gradient vector field of information entropy. Define the UAV as a node and construct a dynamic category graph. The nodes of the dynamic category graph are mapped to the popular points of the Riemannian manifold, the UAV coordinates are converted into grid indices, the mission intensity impact of the UAV is calculated, the metric tensor is calculated, positive definite regularization is added to the metric tensor, the objective function is defined, the Riemann gradient of the objective function is calculated, and the manifold points are updated using exponential mapping to obtain the updated spatial coordinates and mission vector. A weighted linear combination is used to generate a control command vector, which is then converted into control commands and executed. Build a visual interface to display control commands; The gradient vector field of the information entropy includes fusing spatial coordinates, velocity, and acceleration data using extended Kalman filtering, clustering the fused data using K-means clustering, recognizing the image using a pre-trained Swin-Transformer model and outputting the corresponding state variables, stacking spatial coordinates, 3D point cloud, velocity, acceleration, clustering results, and state labels using tensor stacking to construct a sparse tensor, summing the non-zero elements of the sparse tensor to calculate the task intensity probability, calculating the information entropy based on the task intensity probability, calculating the partial derivatives of each dimension in the sparse tensor using the finite difference method based on the information entropy, and combining the partial derivatives using vector synthesis to form the gradient vector field of the information entropy. The dynamic category graph includes defining UAVs as nodes, defining control states greater than 0 as edges, constructing an initial category graph, setting a change threshold using statistical analysis, calculating the Euclidean distance between the spatial position of each UAV and the previous time, filtering out Euclidean distances greater than the change threshold, marking them as affected control states, and recalculating the control states; otherwise, retaining the control states from the previous time, and generating a dynamic category graph. The computation of the metric tensor includes using a grid mapping table to convert UAV coordinates into grid indices and calculating the UAV's mission intensity impact, using the following formula: ， in To determine the impact of the mission intensity of the l-th drone at time t, Let be the task strength probability of the sparse tensor at time t. Let i be the spatial coordinates of the l-th UAV at time t, i be the index of the mission type dimension, j be the index of the spatial coordinates, and k be the index of the state variable. The formula for calculating the metric tensor is: ， in For popular points The metric tensor components at point Q and R are the row and column indices of the manifold dimension, respectively. Euclidean measurement, Let be the information entropy of the sparse tensor at time t. At the location of the drone gradient vector, It is a very small constant; The objective function includes the construction using a weighted objective function, and the objective function is defined as follows: ， in Let be the objective function. Let l be the popular point at time t. Let be the information entropy of the sparse tensor at time t. At the location of the drone spatial coordinates, Let be the geodesic distance between the flow points at position l and position u on the manifold. Let u be the popular point at time t; The calculation of the Riemann gradient of the objective function includes combining the gradient vector and the metric tensor to calculate the Riemann gradient of the objective function, as shown in the formula: ， in For the objective function J at the popular point Riemann gradient at that point Let time t be the l-th drone up to the uth drone Controlled state emission; The updating of manifold points includes updating manifold points using an exponential map based on the Riemann gradient, as shown in the formula: ， in For the updated manifold points, Step size; The control command vector includes calculating the position increment using vector subtraction and generating the control command vector using a weighted linear combination, as shown in the formula: ， in Let be the command vector of the UAV at time t, For task vector weights, This is the updated task vector.

2. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 1, characterized in that: The process of collecting multi-source data, constructing a sparse tensor, defining UAVs as nodes, and building a dynamic category graph includes: Use smart sensors to collect multi-source data from drone swarms and perform noise reduction and normalization processing. The multi-source data includes images, 3D point clouds, spatial coordinates, velocity, and acceleration data; Extended Kalman filtering is used to fuse spatial coordinates, velocity, and acceleration data. K-means clustering is used to cluster the fused data. A pre-trained Swin-Transformer model is used to recognize the image and output the corresponding state variables. Spatial coordinates, 3D point cloud, velocity, acceleration, clustering results, and state labels are stacked using tensor stacking to construct a sparse tensor. The non-zero elements of the sparse tensor are summed to calculate the task intensity probability. The information entropy is calculated based on the task intensity probability. The partial derivatives of each dimension in the sparse tensor are calculated using the finite difference method based on the information entropy. The partial derivatives are combined using vector synthesis to form the gradient vector field of the information entropy. Extract the target grid coordinates from the sparse tensor, convert the grid coordinates into actual geographic coordinates, calculate the direction vector from the UAV to the target location, normalize the direction vector, and obtain the task vector. Calculate the control mapping function between UAVs, set the filtering threshold using the fixed threshold method, and define the filtered control mapping function as a control state mapping function. The drones are defined as nodes, and control states with a value greater than 0 are defined as edges. An initial category graph is constructed. A change threshold is set using statistical analysis. The Euclidean distance between the spatial position of each drone and the previous time is calculated. Euclidean distances greater than the change threshold are filtered out and marked as affected control states. The control states are then recalculated. Otherwise, the control states from the previous time are retained, and a dynamic category graph is generated.

3. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 2, characterized in that: The process of mapping nodes of the dynamic category graph to popular points of the Riemannian manifold, converting UAV coordinates to grid indices, and defining an objective function includes: Based on drones, a Riemannian manifold is constructed. The Euclidean distance between points of the manifold is calculated using the Euclidean distance formula and defined as the Euclidean metric. The initial metric tensor is set using the Euclidean metric, and the nodes of the dynamic category graph are mapped to the manifold points of the Riemannian manifold. We construct the adjacency matrix on the manifold by using the edges of the dynamic category graph as adjacency constraints. Using a grid mapping table, the drone coordinates are converted into grid indices, the mission intensity impact of the drone is calculated, the metric tensor is calculated, and positive definite regularization is applied to the metric tensor. We construct a weighted objective function and define the objective function.

4. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 3, characterized in that: The calculation of the Riemann gradient of the objective function, updating the manifold points using an exponential mapping, and obtaining the updated spatial coordinates and task vector include: The gradient vector and metric tensor are combined to compute the Riemann gradient of the objective function, and the manifold points are updated using an exponential mapping based on the Riemann gradient. Use vector decomposition to decompose the updated spatial coordinates and task vectors from the updated manifold points.

5. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 4, characterized in that: The generation of control command vectors using weighted linear combination includes: The position increment is calculated using vector subtraction, and the control command vector is generated using a weighted linear combination and then normalized.

6. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 5, characterized in that: The process of converting control command vectors into control commands and executing them includes: The normalized control command vector is converted into control commands using a vector set encapsulation method. Extract the adjacency matrix of the dynamic category graph, allocate communication channels for the UAV using a wireless communication protocol, send control commands to the corresponding UAV, and the UAV receives and executes the control commands.

7. The intelligent cooperative control method based on unmanned aerial vehicle (UAV) swarms as described in claim 6, characterized in that: The command to construct the visual interface for displaying control instructions includes: The front-end framework React.js is used to build a visual interface to visualize the control commands obtained from the analysis and the multi-source data collected. Users who have completed real-name verification are allowed to view this information.

8. An intelligent collaborative control system based on unmanned aerial vehicle (UAV) swarms, based on the intelligent collaborative control method based on UAV swarms as described in any one of claims 1 to 7, characterized in that: include, The graph collection module is used to collect and preprocess multi-source data, construct a sparse tensor, calculate the partial derivative of each dimension in the sparse tensor using the finite difference method, combine the partial derivatives using the vector synthesis method as the gradient vector field of information entropy, define the UAV as a node, and construct a dynamic category graph. The target update module is used to map the nodes of the dynamic category graph to the popular points of the Riemannian manifold, convert the UAV coordinates into grid indices, calculate the mission intensity impact of the UAV, calculate the metric tensor, add a regularization term to the metric tensor with positive definite regularization, define the objective function, calculate the Riemann gradient of the objective function, update the manifold points using exponential mapping, and obtain the updated spatial coordinates and mission vector. The instruction execution module is used to generate control instruction vectors using weighted linear combinations, convert the control instruction vectors into control instructions, and execute them. The visualization module is used to build a visual interface to display control commands.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent cooperative control method based on unmanned aerial vehicle swarms as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent cooperative control method based on unmanned aerial vehicle swarms as described in any one of claims 1 to 7.

Images