A three-dimensional trajectory mission planning method for UAV swarm

Abstract

The present invention discloses a method for three-dimensional trajectory mission planning of a drone swarm, comprising step S1: constructing a knowledge graph according to a graph neural network, wherein the nodes of the knowledge graph are the information of each drone, and the edges are the relationships between each drone; step S2: based on the constructed knowledge graph, using an improved multi-objective genetic algorithm to realize the Pareto solution set of multiple objectives, i.e., the trajectory planning of the next state of the drone, and then designing the weights between the multiple objectives according to user needs, so as to obtain the optimal solution, and finally the drone swarm runs according to the result of the multi-objective optimization, and updates the knowledge graph. The present invention discloses a method for three-dimensional trajectory mission planning of a drone swarm, through the knowledge graph constructed by the graph neural network and the multi-objective optimization algorithm, the system can realize more intelligent trajectory planning.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) swarm trajectory planning, and in particular relates to a three-dimensional trajectory mission planning method for an unmanned aerial vehicle (UAV) swarm. Background Art

[0002] With the development of science and technology, drones have been widely used in many fields such as agriculture, power inspection, and traffic monitoring. In actual use, drones often need to fly autonomously in a changing environment and predict the future environment to make the best decision. Therefore, the path planning problem of drones has become an important topic in the research of drone-related technologies.

[0003] Existing drone path or trajectory planning cannot find the optimal multi-target, and often cannot obtain satisfactory results when faced with environmental changes and real-time requirements. In addition, when the number of drones increases and forms a drone cluster, the path planning problem becomes more complicated and requires a more efficient and intelligent method to solve. Summary of the invention

[0004] The main purpose of the present invention is to provide a three-dimensional trajectory mission planning method for a drone swarm. Through the knowledge graph constructed by the graph neural network and the multi-objective optimization algorithm (improved NSGA-Ⅱ), the system can achieve more intelligent trajectory planning.

[0005] To achieve the above objectives, the present invention provides a method for planning a three-dimensional trajectory mission of a drone swarm, comprising the following steps:

[0006] Step S1: construct a knowledge graph based on the graph neural network. The nodes of the knowledge graph are the information of each drone (three-dimensional position, status information: ID, fuel level, etc.), and the edges are the relationships between drones (the weight of the edge is: the distance between two nodes, the energy consumption and time required to reach the two nodes);

[0007] Step S2: Based on the knowledge graph that has been constructed, the improved multi-objective genetic algorithm (NSGA-Ⅱ algorithm) is used to realize the Pareto solution set of multiple objectives (shortest time consumption, minimum risk and shortest distance of the drone swarm), that is, the trajectory planning of the next state of the drone, and then the weights between multiple objectives are designed according to user needs to obtain the optimal solution. Finally, the drone swarm runs according to the results of multi-objective optimization and updates the knowledge graph.

[0008] As a further preferred technical solution of the above technical solution, step S1 is specifically implemented as follows:

[0009] Step S1.1: The graph neural network describes the nodes (each node represents a drone, and the node information includes: the drone ID, three-dimensional spatial position (latitude, longitude and altitude), task completion time, obstacle position: fixed update (buildings, trees, etc.) and real-time updated obstacles (other aircraft, etc.)), and after continuous node status updates, obtains a state with neighbor node information and graph topology characteristics, and finally outputs these nodes to obtain the required results. The first formula represents the update method of the node state, and the second formula represents the output of the updated node state:

[0010] First formula: h V =f(X V ,X CO[V] ,h ne[V] ,X ne[V] );

[0011] The second formula: V =g(h V ,X V );

[0012] Where: f() is the local conversion function; g() is the local output function; o represents the output result; h V Indicates the current embedding result of the node; X V represents the characteristics of node v; X CO[V] Represents the attribute of the edge corresponding to node v; h ne[V] Indicates the state of node v before updating; X ne[V] Represents the characteristics of the neighbor nodes of node v;

[0013] Step S1.2: The first formula and the second formula both represent operations performed on a node as a unit. When the above operations are performed on all nodes, they are optimized to be described by the third formula and the fourth formula:

[0014] The third formula: H = F (H, X);

[0015] The fourth formula: O = G (H, X N );

[0016] Among them, H represents the state of all nodes; O represents the result after outputting all nodes; X represents the characteristics of the edge; X N Respectively represent the characteristics of all nodes; F() and G() represent the global conversion function and the global output function respectively. It can be seen that when the states of all nodes are updated from round t to round t+1, it is expressed as the fifth formula:

[0017] Fifth formula: H t+1 =F(H t ,X);

[0018] Step S1.3: Combine pruning with Graph Neural Networks (GNNs) to construct a knowledge graph. (A knowledge graph is a graph structure that represents entities and their relationships, and a graph neural network can effectively learn the representation of nodes and edges in a graph structure. Pruning can help simplify the graph structure, reduce the size of the graph, and improve learning efficiency.) The pruning strategy is: when the number of edges of a drone (i.e., a node) exceeds a certain threshold (3-5, which can be set according to the size of the drone swarm), remove the edges whose edge weights are greater than a certain threshold, and then retain important connections by adjusting the threshold.

[0019] As a further preferred technical solution of the above technical solution, step S2 is specifically implemented as the following steps:

[0020] Step S2.1: Assume that the current position of the drone group is the initial position, and the final position of the drone group is the target position. The multi-objective genetic algorithm is used to achieve the multi-objectives of the drone group, including the shortest time consumption, the shortest distance, the least energy consumption and the least risk (Euclidean distance between two drones);

[0021] The constraint function is obstacles: fixed updates (buildings, trees, etc.) and real-time updates (other aircraft, etc.) and set them as no-fly zones;

[0022] Step S2.2: Use adaptive parameters to improve the multi-objective genetic algorithm, give lower crossover rate Pc and mutation rate Pm values ​​to high fitness individuals for preservation; give high Pc and Pm values ​​to low fitness individuals for the change of inferior individuals;

[0023]

[0024] fm is the optimal fitness value of the population, fa is the average fitness value of the population, f′ is the fitness value with better fitness among the two individuals to be crossed, f is the fitness value of the individual to be mutated, k1, k2, k3, k4 are all adaptive control parameters, and take fixed values ​​in the interval (0,1).

[0025] To achieve the above objectives, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the three-dimensional trajectory mission planning method for the drone swarm when executing the program.

[0026] To achieve the above objectives, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the three-dimensional trajectory mission planning method of the drone swarm are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of a multi-objective genetic algorithm of a three-dimensional trajectory mission planning method for a drone swarm of the present invention. DETAILED DESCRIPTION

[0028] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0029] In a preferred embodiment of the present invention, those skilled in the art should note that the drone swarm and the like involved in the present invention may be regarded as prior art.

[0030] Preferred embodiments.

[0031] The present invention discloses a method for planning a three-dimensional trajectory mission of a drone group, comprising the following steps:

[0032] Step S1: construct a knowledge graph based on the graph neural network. The nodes of the knowledge graph are the information of each drone (three-dimensional position, status information: ID, fuel level, etc.), and the edges are the relationships between drones (the weight of the edge is: the distance between two nodes, the energy consumption and time required to reach the two nodes);

[0033] Step S2: Based on the knowledge graph that has been constructed, the improved multi-objective genetic algorithm (NSGA-Ⅱ algorithm) is used to realize the Pareto solution set of multiple objectives (shortest time consumption, minimum risk and shortest distance of the drone swarm), that is, the trajectory planning of the next state of the drone, and then the weights between multiple objectives are designed according to user needs to obtain the optimal solution. Finally, the drone swarm runs according to the results of multi-objective optimization and updates the knowledge graph.

[0034] Specifically, step S1 is implemented as follows:

[0035] Step S1.1: The graph neural network describes the nodes (each node represents a drone, and the node information includes: the drone ID, three-dimensional spatial position (latitude, longitude and altitude), task completion time, obstacle position: fixed update (buildings, trees, etc.) and real-time updated obstacles (other aircraft, etc.)), and after continuous node status updates, obtains a state with neighbor node information and graph topology characteristics, and finally outputs these nodes to obtain the required results. The first formula represents the update method of the node state, and the second formula represents the output of the updated node state:

[0036] First formula: h V =f(XV ,X CO[V] ,h ne[V] ,X ne[V] );

[0037] The second formula: V =g(h V ,X V );

[0038] Where: f() is the local conversion function; g() is the local output function; o represents the output result; h V Indicates the current embedding result of the node; X V represents the characteristics of node v; X CO[V] Represents the attribute of the edge corresponding to node v; h ne[V] Indicates the state of node v before updating; X ne[V] Represents the characteristics of the neighbor nodes of node v;

[0039] Step S1.2: The first formula and the second formula both represent operations performed on a node as a unit. When the above operations are performed on all nodes, they are optimized to be described by the third formula and the fourth formula:

[0040] The third formula: H = F (H, X);

[0041] The fourth formula: O = G (H, X N );

[0042] Among them, H represents the state of all nodes; O represents the result after outputting all nodes; X represents the characteristics of the edge; X N Respectively represent the characteristics of all nodes; F() and G() represent the global conversion function and the global output function respectively. It can be seen that when the states of all nodes are updated from round t to round t+1, it is expressed as the fifth formula:

[0043] Fifth formula: H t+1 =F(H t ,X);

[0044] Step S1.3: Combine pruning with Graph Neural Networks (GNNs) to construct a knowledge graph. (A knowledge graph is a graph structure that represents entities and their relationships, and a graph neural network can effectively learn the representation of nodes and edges in a graph structure. Pruning can help simplify the graph structure, reduce the size of the graph, and improve learning efficiency.) The pruning strategy is: when the number of edges of a drone (i.e., a node) exceeds a certain threshold (3-5, which can be set according to the size of the drone swarm), remove the edges whose edge weights are greater than a certain threshold, and then retain important connections by adjusting the threshold.

[0045] More specifically, step S2 is implemented as the following steps (improved multi-objective genetic algorithm):

[0046] Step S2.1: Assume that the current position of the drone group is the initial position, and the final position of the drone group is the target position. The multi-objective genetic algorithm is used to achieve the multi-objectives of the drone group, including the shortest time consumption, the shortest distance, the least energy consumption and the least risk (Euclidean distance between two drones);

[0047] The constraint function is obstacles: fixed updates (buildings, trees, etc.) and real-time updates (other aircraft, etc.) and set them as no-fly zones;

[0048] Step S2.2: Use adaptive parameters to improve the multi-objective genetic algorithm, give lower crossover rate Pc and mutation rate Pm values ​​to high fitness individuals for preservation; give high Pc and Pm values ​​to low fitness individuals for the change of inferior individuals;

[0049]

[0050]

[0051] fm is the optimal fitness value of the population, fa is the average fitness value of the population, f′ is the fitness value with better fitness among the two individuals to be crossed, f is the fitness value of the individual to be mutated, k1, k2, k3, k4 are all adaptive control parameters, and take fixed values ​​in the interval (0,1).

[0052] By calculating the fitness value (such as the shortest distance) through initialization, mutation crossover, etc., the optimal fitness value can be found.

[0053] The specific steps of the multi-objective genetic algorithm are:

[0054] 1. Population initialization;

[0055] 2. Calculate the fitness value based on at least three objective functions (including the shortest time spent by the drone group, the least energy consumption, the lowest risk (need to detour if there is a roadblock) and the shortest distance);

[0056] 3. Fast non-dominated sorting and congestion calculation;

[0057] 4. Adaptive tournament selects appropriate parents;

[0058] 5. Crossover mutation produces offspring and calculates fitness value;

[0059] 6. Fast non-dominated sorting and congestion calculation;

[0060] 7. Select the first N individuals;

[0061] 8. Determine whether the number of iterations has been reached. If so, end the process. Otherwise, continue the adaptive tournament to select a suitable parent.

[0062] The final output value after reaching the number of iterations is the ranking of multiple sets of optimal solutions of the multi-objective function, and then a selection is made according to one's own needs (for example, if the shortest distance is required, the set of optimal solution data with the shortest distance can be selected).

[0063] Preferably, the present invention also discloses an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for three-dimensional trajectory mission planning of a swarm of unmanned aerial vehicles when executing the program.

[0064] Preferably, the present invention also discloses a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for three-dimensional trajectory mission planning of a swarm of unmanned aerial vehicles.

[0065] The advantages of the present invention are:

[0066] (1) Intelligent trajectory planning: Through the knowledge graph constructed by graph neural network and multi-objective optimization algorithm (improved NSGA-Ⅱ), the system can achieve more intelligent trajectory planning.

[0067] (2) Multi-objective optimization: Using the multi-objective optimization algorithm, the system can comprehensively consider multiple objectives in trajectory planning, such as the shortest path, minimum risk, minimum energy consumption, etc., to achieve the purpose of balance and optimization.

[0068] (3) Strong adaptability: Through graph neural network learning and dynamic updating of knowledge graphs, the system can adapt to different environmental conditions and task requirements, improving the adaptability of trajectory planning.

[0069] (4) Efficient flight: The optimized trajectory planning algorithm is expected to achieve more efficient flight, reduce flight time, reduce energy consumption, and improve the overall work efficiency of the drone swarm.

[0070] (5) Real-time response: The system is real-time and uses cloud computing to adjust the trajectory of the drone in real time during flight to adapt to changes in the dynamic environment, reduce collision risks, and improve safety.

[0071] (6) Collaborative flight support: The collaborative flight of UAVs, the system supports coordination and communication between large-scale aircraft to achieve more complex tasks.

[0072] (7) Task Customization: The system has a certain degree of flexibility and can be customized according to the requirements of specific tasks, thus adapting to different application scenarios and task types.

[0073] (8) Reduce human intervention: Through the automated trajectory planning process, the dependence on operators is reduced and the autonomy of the drone swarm is improved.

[0074] (9) Uncertainty handling: Uncertainties in a dynamic environment, such as weather changes, air traffic, etc., may cause the planned path to be inapplicable during execution. The system can handle these uncertainties effectively.

[0075] It is worth mentioning that the technical features such as drone swarms involved in the patent application of this invention should be regarded as prior art. The specific structure, working principle, and possible control method and spatial layout method of these technical features can be selected according to the conventional methods in the field, and should not be regarded as the inventive point of this patent. This patent will not be further elaborated.

[0076] For those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A three-dimensional trajectory mission planning method for a drone swarm, characterized in that: The following steps are involved: Step S1: construct a knowledge graph based on the graph neural network, where the nodes of the knowledge graph are the information of each drone, and the edges are the relationships between the drones; Step S2: Based on the knowledge graph that has been constructed, the improved multi-objective genetic algorithm is used to realize the Pareto solution set of multiple objectives, that is, the trajectory planning of the next state of the drone, and then the weights between the multiple objectives are designed according to user needs to obtain the optimal solution. Finally, the drone swarm runs according to the results of the multi-objective optimization and updates the knowledge graph; Step S1 is specifically implemented as follows: Step S1.1: The graph neural network describes the nodes, and after continuous node status updates, obtains a state with neighbor node information and graph topology characteristics, and finally outputs these nodes to obtain the required results. The first formula represents the update method of the node status, and the second formula represents the output of the updated node status: First formula: h V =f(X V , X CO[V] ,h ne[V] , X ne[V] ); The second formula: V =g(h V , X V ); Where: f( ) is the local conversion function; g( ) is the local output function; o represents the output result; h V Indicates the current embedding result of the node; X V represents the characteristics of node v; X CO[V] Represents the attribute of the edge corresponding to node v; h ne[V] Indicates the state of node v before updating; X ne[V] Represents the characteristics of the neighbor nodes of node v; Step S1.2: The first formula and the second formula both represent operations performed on a node as a unit. When the above operations are performed on all nodes, they are optimized to be described by the third formula and the fourth formula: The third formula: H = F (H, X); The fourth formula: O = G (H, X N ); Among them, H represents the state of all nodes; 0 represents the result after outputting all nodes; X represents the characteristics of the edge; X N Respectively represent the characteristics of all nodes; F( ) and G( ) respectively represent the global conversion function and the global output function. It can be seen that when the states of all nodes are updated from round t to round t+1, it is expressed as the fifth formula: Fifth formula: H t+1 =F(H t , X); Step S1.3: Pruning is combined with graph neural network to construct a knowledge graph. The pruning strategy is: when the number of edges of a drone exceeds a certain threshold, the edges with edge weights greater than a certain threshold are removed, and then the threshold is adjusted to retain important connections; The improved multi-objective genetic algorithm of step S2 is specifically implemented as follows: Step S2.1: Assume that the current position of the drone group is the initial position, and the final position of the drone group is the target position. The multi-objective genetic algorithm is used to achieve the multiple goals of the drone group, including the shortest time consumption, the shortest distance, the least energy consumption and the least risk. The constraint function is obstacles: fixed updates and real-time updates of obstacles, and setting them as no-fly zones; Step S2.2: Use adaptive parameters to improve the multi-objective genetic algorithm, give lower crossover rate Pc and mutation rate Pm values ​​to high fitness individuals for preservation; give high Pc and Pm values ​​to low fitness individuals for the change of inferior individuals; fm is the optimal fitness value of the population, fa is the average fitness value of the population, f′ is the fitness value with better fitness among the two individuals to be crossed, f is the fitness value of the individual to be mutated, k1, k2, k3, k4 are all adaptive control parameters, and take fixed values ​​in the interval (0,1).

2. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the drone swarm three-dimensional trajectory mission planning method as claimed in claim 1 are implemented.

3. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for three-dimensional trajectory mission planning of a drone swarm as described in claim 1 are implemented.

Images