Intelligent cooperative control method for unmanned aerial vehicle group

Abstract

This invention relates to the field of unmanned aerial vehicle (UAV) swarm control, and more particularly to an intelligent cooperative control method for UAV swarms. The method includes: receiving and preprocessing environmental data, UAV status data, and task data; performing data fusion processing on the preprocessed data; using a multi-objective optimization genetic algorithm to allocate tasks to the UAV swarm and generate a task allocation scheme; generating an optimal path based on the task allocation scheme; the UAVs executing tasks according to the optimal path, adjusting their flight paths using a dynamic obstacle avoidance function; generating a new optimal path based on the UAVs' current positions; the UAVs continuing to execute tasks according to the new optimal path; and returning to the landing point after completing the task and uploading collected flight and environmental data. This method solves the problems that existing selection mechanisms often rely too heavily on a single criterion, existing path planning methods mostly only consider the static environment, and the lack of real-time obstacle avoidance capabilities that lead to collisions or path deviations during task execution.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control, and more particularly to a method for intelligent collaborative control of UAV swarms. Background Technology

[0002] With the rapid development of drone technology, drone swarms have been widely used in various fields, such as logistics and delivery, agricultural management, inspection and monitoring, and disaster relief. Drone swarm cooperative control technology plays a crucial role in achieving efficient collaborative operations among multiple drones. Currently, researchers have developed various methods and algorithms to optimize task allocation, path planning, and dynamic obstacle avoidance for drone swarms, thereby improving their overall efficiency and safety. Among these, optimization techniques such as genetic algorithms, ant colony optimization, and particle swarm optimization are widely used for task allocation and path planning in drone swarms, while dynamic obstacle avoidance technology based on sensor data has also been extensively researched and applied.

[0003] Despite the progress made in intelligent collaborative control of drone swarms, existing drone technologies still face challenges in three key areas: task allocation, path planning, and dynamic obstacle avoidance. In terms of task allocation, methods struggle to simultaneously consider task completion time, power consumption, and task complexity, exhibiting weak adaptability to dynamic changes and environmental uncertainties. Regarding path planning, traditional algorithms perform well in static environments but suffer from low efficiency and insufficient reliability in dynamic environments. As for dynamic obstacle avoidance, reliance on sensor data leads to poor adaptability when facing various dynamic obstacles, resulting in delayed responses or excessive avoidance, causing detours in flight paths, increasing task completion time and energy consumption, and exhibiting insufficient real-time performance and robustness. Summary of the Invention

[0004] This invention proposes an intelligent collaborative control method for UAV swarms, addressing the problems of existing selection mechanisms that often rely too heavily on a single criterion, leading to insufficient swarm diversity and affecting the search for the global optimal solution; existing path planning methods that mostly consider only static environments and lack real-time obstacle avoidance capabilities in dynamic environments; in complex real-world application scenarios, UAVs need to face dynamic factors such as moving obstacles, and traditional static path planning methods are ill-equipped to meet these challenges; and the lack of real-time obstacle avoidance capabilities that can cause UAVs to easily collide or deviate from their paths during mission execution, severely impacting mission safety and efficiency.

[0005] The intelligent collaborative control method for unmanned aerial vehicle (UAV) swarms includes the following steps:

[0006] S1: Receive and preprocess environmental data, UAV status data, and task data to obtain preprocessed data, and perform data fusion processing on the preprocessed data to obtain fused data; use a multi-objective optimization genetic algorithm to allocate tasks to the UAV swarm and generate a task allocation scheme.

[0007] S2: Based on the task allocation scheme, an optimal path is generated using an optimal path planning algorithm based on graph neural networks; the UAV executes the task according to the optimal path and adjusts the UAV's flight path using a dynamic obstacle avoidance function; the node information in the graph structure is updated according to the UAV's current position, and a new optimal path is generated using an optimal path planning algorithm based on graph neural networks; the UAV continues to execute the task according to the new optimal path, and returns to the landing point after the task is completed and uploads the collected flight data and environmental data.

[0008] Preferably, S1 specifically includes:

[0009] In the implementation of the multi-objective optimization genetic algorithm, population initialization is performed to generate an initial population, and an initial task assignment scheme for each individual in the population is randomly generated; the initial task assignment scheme is evaluated and optimized by calculating the fitness value.

[0010] Preferably, S1 specifically includes:

[0011] In the process of allocating tasks for a drone swarm using a multi-objective optimization genetic algorithm, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​to enter the mating pool.

[0012] Preferably, S1 specifically includes:

[0013] Based on the selected individuals, single-point crossover and random mutation operations are performed to obtain the next generation population. The fitness value of each individual in the next generation population is calculated, and a fast non-dominated sort is used to obtain the Pareto front solution. Based on the fitness value, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​from the Pareto front solution for single-point crossover and random mutation operations to generate a new population. Through iterative operations, the optimized Pareto front solution is output, and a task assignment scheme is selected from the optimized Pareto front solution.

[0014] Preferably, S2 specifically includes:

[0015] The optimal path planning algorithm based on graph neural networks combines graph neural networks, convolutional neural networks, and fully connected neural networks to improve feature extraction and path planning decisions.

[0016] Preferably, S2 specifically includes:

[0017] In the process of calculating the optimal path using the graph neural network-based optimal path planning algorithm, a graph structure is established for each UAV based on the task allocation scheme, and the initial features of the UAV graph structure are extracted using the graph neural network.

[0018] Preferably, S2 specifically includes:

[0019] The initial features are input into a convolutional neural network for processing to obtain the features processed by the convolutional neural network; the features processed by the convolutional neural network are then input into a fully connected neural network for path planning and decision-making to obtain the optimal path.

[0020] Preferably, S2 specifically includes:

[0021] When the drone executes its mission along the optimal path, it detects obstacles in the surrounding area in real time. When there are obstacles, it uses a dynamic obstacle avoidance function to calculate the obstacle avoidance vector and determine the direction and distance of obstacle avoidance. Based on the obstacle avoidance vector, it adjusts the drone's flight direction and position, flies in the calculated direction, and moves to the indicated position.

[0022] Preferably, S2 specifically includes:

[0023] The formula for the dynamic obstacle avoidance function is as follows:

[0024]

[0025] in, α is the obstacle avoidance vector; M is the obstacle avoidance coefficient; w is the total number of obstacles currently detected. γ d is the dynamic weight of the γth obstacle; γ It is the current distance between the drone and the γth obstacle; It is the direction vector of the current drone to the γth obstacle, indicating the direction of obstacle avoidance.

[0026] The beneficial effects of the technical solution in the embodiments of the present invention are:

[0027] 1. The multi-objective optimization genetic algorithm improves the efficiency and accuracy of task allocation for UAV swarms by optimizing fitness values ​​and selection mechanisms, combined with an improved roulette wheel algorithm. It enhances the diversity and global search capabilities of the population, ensuring that each UAV can be reasonably assigned tasks, maximizing task completion efficiency and resource utilization.

[0028] 2. By using graph neural networks to extract initial features and combining them with convolutional neural networks and fully connected neural networks for processing and path planning decisions, the path planning problem in complex environments is effectively handled, and the globality and accuracy of path planning are improved.

[0029] 3. By combining real-time sensor data, a dynamic obstacle avoidance function is defined. Based on the real-time position and dynamic changes of obstacles, the flight path of the UAV is flexibly adjusted to ensure the accuracy and timeliness of obstacle avoidance, thereby improving the safety and stability of the UAV when performing tasks. Attached Figure Description

[0030] Figure 1 This is a flowchart of the intelligent collaborative control method for unmanned aerial vehicle swarms described in this invention. Detailed Implementation

[0031] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] The specific scheme of the intelligent collaborative control method for unmanned aerial vehicle swarms provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0034] See attached document Figure 1 The diagram illustrates a flowchart of an intelligent collaborative control method for unmanned aerial vehicle (UAV) swarms provided by an embodiment of the present invention. The method includes the following steps:

[0035] S1: Receive and preprocess environmental data, UAV status data, and task data to obtain preprocessed data, and perform data fusion processing on the preprocessed data to obtain fused data; use a multi-objective optimization genetic algorithm to allocate tasks to the UAV swarm and generate a task allocation scheme.

[0036] First, it receives environmental data from control center personnel, such as geographic information and obstacle locations; UAV status data, such as the number of UAVs, real-time location coordinates, flight speed, flight altitude, power consumption rate, flight trajectory, and UAV performance parameters; and task data, such as the number of tasks, task target location, task complexity, task completion time, and task priority. The types of data provided by control center personnel are set according to the specific implementation scenario.

[0037] Secondly, environmental data, UAV status data, and mission data are preprocessed. Preprocessing includes data cleaning and format conversion to obtain preprocessed data. The preprocessed data is then fused using a weighted average fusion technique to obtain the fused data. Preprocessing and weighted average fusion techniques are existing technologies and will not be elaborated upon here.

[0038] Finally, a multi-objective optimization genetic algorithm is used to perform initial task allocation on the fused data to determine the specific task of each UAV. The multi-objective optimization genetic algorithm improves the efficiency and accuracy of task allocation for UAV swarms by defining and optimizing fitness values ​​and a selection mechanism based on an improved roulette wheel algorithm, enhances the diversity and global search capability of the population, and ensures that each UAV can be reasonably allocated tasks to maximize the efficiency of task completion and resource utilization.

[0039] In multi-objective optimization genetic algorithms, a population contains multiple individuals, each representing a specific UAV task allocation scheme. The UAV task allocation scheme is a specific strategy for assigning tasks to corresponding UAVs, usually represented as a vector or array. The position of each element in the vector represents the task number, and the value of each element represents the UAV number that performs the corresponding task.

[0040] The process of using a multi-objective optimization genetic algorithm for task allocation in a drone swarm is as follows:

[0041] First, population initialization is performed. An initial population is generated using a combination of random and heuristic generation methods. Based on the number of drones and the number of missions, an initial mission allocation scheme for each individual in the population is randomly generated. The population initialization method is existing technology; the specific method will be set according to the specific implementation scenario.

[0042] Then, the initial task allocation scheme is evaluated and optimized by calculating the fitness value. The fitness value combines task completion time, power consumption, and task complexity to evaluate the initial task allocation scheme for each individual. The fitness value is defined as:

[0043]

[0044] Where f is the fitness value of individual i, used to evaluate the quality of the initial task assignment scheme for individual i; T i It is the task completion time of individual i, obtained by summing the time of individual i in completing all assigned tasks; E i This represents the power consumption of individual i, indicating the total power required for individual i to complete the task. It is obtained by summing the power consumption of individual i in completing all assigned tasks. Power consumption can be estimated based on existing UAV power consumption models, which will not be elaborated here; C iT is the task complexity of individual i, obtained by summing the task complexities of all tasks assigned to individual i; max and T min These are the maximum and minimum task completion times for all individuals; E max and E min These are the maximum and minimum power consumption values ​​for all individuals; w1, w2, and w3 are the weights of task completion time, power consumption, and task complexity, used to balance the importance of task completion time, power consumption, and task complexity in the fitness value, and are set according to the specific application scenario.

[0045] Furthermore, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​for the mating pool. The formula for the improved roulette wheel algorithm is as follows:

[0046]

[0047] Where, p i f is the probability that individual i is selected; y It is the fitness value of individual y; It is the total number of individuals in the population; the improved roulette wheel betting algorithm uses logarithmic operations to avoid numerical overflow, while ensuring the numerical stability of the improved roulette wheel betting algorithm formula.

[0048] Next, a single-point crossover operation is used to cross-pair the selected individuals to generate new offspring. A random mutation operation is then performed on the newly generated offspring to increase population diversity, avoid getting trapped in local optima, and obtain the next generation population. The fitness value of each individual in the next generation population is recalculated, and a fast non-dominated sorting algorithm is used to obtain the Pareto front solution, which is a set of non-dominated optimal solutions in a multi-objective optimization problem. Further, based on the fitness value, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​from the Pareto front solution for single-point crossover and random mutation operations to generate a new population. The above steps are repeated, iterating multiple times to continuously optimize the population. The iteration terminates when the maximum number of iterations is reached or the change in the fitness value of the new population is less than a preset threshold. The maximum number of iterations and the preset threshold are set according to the specific implementation scenario. The fast non-dominated sorting, single-point crossover, and random mutation techniques are all existing technologies and will not be elaborated upon here.

[0049] The output of the multi-objective optimization genetic algorithm is a set of optimized Pareto front solutions. Decision-makers select the solution that best meets the current needs from the optimized Pareto front solutions based on actual requirements and specific application scenarios, and use it as the task allocation scheme.

[0050] S2: Based on the task allocation scheme, an optimal path is generated using an optimal path planning algorithm based on graph neural networks; the UAV executes the task according to the optimal path and adjusts the UAV's flight path using a dynamic obstacle avoidance function; the node information in the graph structure is updated according to the UAV's current position, and a new optimal path is generated using an optimal path planning algorithm based on graph neural networks; the UAV continues to execute the task according to the new optimal path, and returns to the landing point after the task is completed and uploads the collected flight data and environmental data.

[0051] The optimal path planning algorithm based on graph neural networks combines three technologies: graph neural networks (GNN), convolutional neural networks (CNN), and fully connected neural networks (FNN). Based on existing technologies such as GNN, CNN, and FNN, it improves the feature extraction and path planning decision-making parts by incorporating the UAV's flight speed, power consumption rate, task priority, and task distance to enhance the accuracy and adaptability of path planning and better meet the needs of UAV missions.

[0052] The process of calculating the optimal path using the graph neural network-based optimal path planning algorithm is as follows:

[0053] First, referring to the task allocation scheme, a graph structure is established for each UAV using expert experience. The nodes of the graph represent the UAV's position in the flight environment during mission execution, including the start point, end point, and intermediate waypoints; the edges represent the flight path, and the edge weights consider mission distance, flight speed, power consumption rate, mission priority, and obstacle information; where mission distance refers to the total distance between the UAV's current starting position and the mission target position. Then, a graph neural network (GNN) is used to extract initial features of the UAV's graph structure, incorporating the corresponding flight speed and power consumption rate of the UAV to make the extracted initial features more closely reflect actual flight conditions. The formula for feature extraction using the graph neural network (GNN) is as follows:

[0054]

[0055] in, The initial features of node v after the (t+1)th iteration are: σ is the activation function; W is the weight matrix of the GNN; e uv The eigenvectors of edge uv; is the initial feature obtained by node u after the t-th iteration; node u is a neighbor node of node v; |N(v)| is the set of neighbor nodes of node v; V is the flight speed of the UAV corresponding to the graph structure; P is the power consumption rate of the UAV corresponding to the graph structure; b is the bias vector of GNN.

[0056] Furthermore, the initial features extracted by the GNN are input into the CNN for further processing, resulting in CNN-processed features. By incorporating remaining battery power and task priority, the consideration of battery power and task importance during local feature processing is enhanced, making path planning more aligned with the drone's endurance and mission urgency. The CNN feature processing formula is as follows:

[0057]

[0058] Among them, c v This represents the features of node v after CNN processing; This is the weight matrix of the CNN; h v These are the initial features of node v; K(v) is the bias vector of the CNN; H(v) is the bias vector of the CNN. v The set of convolutional neighborhoods in CNNs; It is the weight matrix of the convolutional neighborhood features in a CNN; It is a node Initial characteristics; nodes It represents the neighboring nodes of node v; F is the remaining battery power of the drone corresponding to the graph structure; U is the task priority of the drone corresponding to the graph structure; F max This is the maximum remaining battery power of all drones; U max It is the highest task priority among all tasks;

[0059] Finally, the features processed by the CNN are input into the FNN for path planning decisions. The FNN integrates global and local features and incorporates task complexity and task distance, allowing path planning to better balance task complexity, flight efficiency, and task distance. The formula for path planning decisions using the FNN is as follows:

[0060]

[0061] Where o is the path planning result of the UAV corresponding to the graph structure, i.e. the optimal path; R is the weight matrix of the FNN output layer; It is the bias vector of the FNN; It is the feature c of node v after CNN processing. v The intermediate feature vector in the l-th layer of the FNN; R lis the weight matrix of the l-th layer of the FNN; L is the number of layers in the fully connected neural network; C is the task complexity of the UAV corresponding to the graph structure; D is the task distance of the UAV corresponding to the graph structure, which is obtained by calculating the sum of the path lengths between the UAV's starting position, each intermediate waypoint and the target position in the graph structure. The distance of each path segment is calculated using the Euclidean distance formula or the Manhattan distance formula, and they are summed to obtain the total distance. The Euclidean distance formula or the Manhattan distance formula is existing technology and will not be elaborated here.

[0062] When a drone executes a mission along the optimal path, it also needs to combine real-time sensor data to define a dynamic obstacle avoidance function. Real-time sensor data includes the distance to obstacles, their direction vectors, dynamic weights, and the total number of obstacles. Based on this real-time sensor data, the dynamic obstacle avoidance function flexibly adjusts the drone's flight path, improving the drone's safety and stability during mission execution. The formula for the dynamic obstacle avoidance function is as follows:

[0063]

[0064] in, α is the obstacle avoidance vector, which helps the drone safely avoid obstacles by providing new directions and distances; α is the obstacle avoidance coefficient, used to adjust the intensity of obstacle avoidance and determining the sensitivity of the drone's position adjustment; M is the total number of obstacles currently detected; w γ This is the dynamic weight of the γth obstacle. The dynamic weight of the obstacle is calculated by combining real-time sensor data with preset rules. The obstacle avoidance coefficient and specific preset rules are set according to the specific implementation scenario; d γ This is the current distance between the drone and the γth obstacle, reflecting the magnitude of the potential collision risk; It is the direction vector of the current drone to the γth obstacle, indicating the direction of obstacle avoidance.

[0065] The drone executes its mission along the optimal path. During mission execution, the drone uses sensors to detect surrounding obstacles in real time. If an obstacle is found, a dynamic obstacle avoidance function is used to calculate the obstacle avoidance vector, determining the direction and distance for obstacle avoidance. The drone adjusts its flight direction and position based on the calculated obstacle avoidance vector, flying in the calculated direction and moving to the indicated position. After obstacle avoidance is completed, the node information in the graph structure is updated based on the drone's current position, and a new optimal path is generated using an optimal path planning algorithm based on a graph neural network. The drone continues to execute its mission along the new optimal path.

[0066] After successfully completing its mission, the drone needs to return to the designated landing point and upload the collected flight data (flight trajectory, flight speed, flight altitude, power consumption rate) and environmental data to the ground control center or cloud server. This data will be analyzed by the data science team at the ground control center or cloud server to evaluate the drone's mission completion and performance.

[0067] In summary, a method for intelligent collaborative control of unmanned aerial vehicle (UAV) swarms has been developed.

[0068] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0069] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0070] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent collaborative control of unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: S1: Receives and preprocesses environmental data, UAV status data, and task data to obtain preprocessed data. Then, it performs data fusion processing on the preprocessed data to obtain fused data. A multi-objective optimization genetic algorithm is used for task allocation in the UAV swarm to generate a task allocation scheme. In the multi-objective optimization genetic algorithm, the fitness value is defined as follows: ， in, Individual fitness value; Individual Task completion time; Individual Power consumption; Individual Task complexity; and These are the maximum and minimum task completion times for all individuals; and These are the maximum and minimum power consumption of all individuals; , , These are the weights of task completion time, power consumption, and task complexity. S2: Based on the task allocation scheme, the optimal path is generated using the optimal path planning algorithm based on graph neural network; the UAV executes the task according to the optimal path and adjusts the flight path of the UAV using the dynamic obstacle avoidance function; the node information in the graph structure is updated according to the current position of the UAV, and a new optimal path is generated using the optimal path planning algorithm based on graph neural network. The UAV continues to execute the task according to the new optimal path, and returns to the landing point after the task is completed and uploads the collected flight data and environmental data. The formula for the dynamic obstacle avoidance function is as follows: ， in, It is the obstacle avoidance vector; It is the obstacle avoidance coefficient; This represents the total number of obstacles currently detected. It is the first Dynamic weights of each obstacle; It is the current drone and the first The distance between the obstacles; Is this the current drone to the 1st The direction vector of each obstacle indicates the direction of obstacle avoidance.

2. The intelligent collaborative control method for unmanned aerial vehicle swarms according to claim 1, characterized in that, S1 specifically includes: In the implementation of the multi-objective optimization genetic algorithm, population initialization is performed to generate an initial population, and an initial task assignment scheme for each individual in the population is randomly generated; the initial task assignment scheme is evaluated and optimized by calculating the fitness value.

3. The intelligent collaborative control method for unmanned aerial vehicle (UAV) swarms according to claim 2, characterized in that, S1 specifically includes: In the process of allocating tasks for a drone swarm using a multi-objective optimization genetic algorithm, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​to enter the mating pool.

4. The intelligent collaborative control method for unmanned aerial vehicle swarms according to claim 3, characterized in that, S1 specifically includes: Based on the selected individuals, single-point crossover and random mutation operations are performed to obtain the next generation population. The fitness value of each individual in the next generation population is calculated, and a fast non-dominated sort is used to obtain the Pareto front solution. Based on the fitness value, an improved roulette wheel algorithm is used to select individuals with high fitness values ​​from the Pareto front solution for single-point crossover and random mutation operations to generate a new population. Through iterative operations, the optimized Pareto front solution is output, and a task assignment scheme is selected from the optimized Pareto front solution.

5. The intelligent collaborative control method for unmanned aerial vehicle swarms according to claim 1, characterized in that, S2 specifically includes: The optimal path planning algorithm based on graph neural networks combines graph neural networks, convolutional neural networks, and fully connected neural networks to improve feature extraction and path planning decisions.

6. The intelligent cooperative control method for unmanned aerial vehicle swarms according to claim 5, characterized in that, S2 specifically includes: In the process of calculating the optimal path using the graph neural network-based optimal path planning algorithm, a graph structure is established for each UAV based on the task allocation scheme, and the initial features of the UAV graph structure are extracted using the graph neural network.

7. The intelligent collaborative control method for unmanned aerial vehicle swarms according to claim 6, characterized in that, S2 specifically includes: The initial features are input into a convolutional neural network for processing to obtain the features processed by the convolutional neural network; the features processed by the convolutional neural network are then input into a fully connected neural network for path planning and decision-making to obtain the optimal path.

8. The intelligent collaborative control method for unmanned aerial vehicle swarms according to claim 7, characterized in that, S2 specifically includes: When the drone executes its mission along the optimal path, it detects obstacles in the surrounding area in real time. When there are obstacles, it uses a dynamic obstacle avoidance function to calculate the obstacle avoidance vector and determine the direction and distance of obstacle avoidance. Based on the obstacle avoidance vector, it adjusts the drone's flight direction and position, flies in the calculated direction, and moves to the indicated position.

Images