Unmanned aerial vehicle formation cooperative task planning method based on hierarchical intelligence

Abstract

The application belongs to the technical field of unmanned aerial vehicle cooperative formation, and specifically discloses a layered intelligent-based unmanned aerial vehicle formation cooperative task planning method, which comprises the following steps: based on an improved deep Q network algorithm, an original formation is split into multiple sub-formations according to formation characteristics and target value information, and target distribution is performed; based on a position matching algorithm of bearing and distance, each unmanned aerial vehicle is smoothly transitioned from a current formation position to a target sub-formation formation position; based on an improved multi-population cooperative evolution algorithm, a cooperative track from a starting point to a target point is planned for each sub-formation, collision detection and track output are performed, and unmanned aerial vehicle formation cooperative task planning is completed. The application solves the problems that the prior art is difficult to adapt to a complex environment, high efficiency, stability and safety cannot be ensured when the original formation is split into sub-formations, and the cooperative nature of a track planned by a traditional genetic algorithm is insufficient.

Description

Technical Field

[0001] This invention belongs to the field of UAV collaborative formation technology, specifically relating to a UAV formation collaborative task planning method based on hierarchical intelligence. Background Technology

[0002] In complex environments, existing UAV formation mission planning methods suffer from the following shortcomings: First, after large UAV formations enter an area, there is a lack of efficient formation splitting and task allocation methods, making it difficult to adapt to complex environments. Second, during formation switching, a smooth position allocation algorithm is needed to ensure efficiency, stability, and safety when splitting the original formation into sub-formations. Finally, in multi-formation collaborative trajectory planning, traditional genetic algorithms are prone to getting trapped in local optima and lack inter-population cooperation mechanisms, resulting in insufficient trajectory planning coordination. These deficiencies limit the autonomy and intelligence level of UAV formations in actual combat.

[0003] Therefore, there is an urgent need for an integrated method that enables autonomous formation splitting, smooth transition, and collaborative trajectory planning to improve the mission execution efficiency of UAV formations in saturation situations. Summary of the Invention

[0004] The purpose of this invention is to address the problems of existing technologies being unable to adapt to complex environments, failing to ensure efficiency, stability, and security when splitting the original formation into sub-formations, and the insufficient coordination of traditional genetic algorithm trajectory planning. This invention proposes a UAV formation collaborative task planning method based on hierarchical intelligence.

[0005] The technical solution of this invention is: a method for collaborative task planning of unmanned aerial vehicle (UAV) formations based on hierarchical intelligence, comprising the following steps:

[0006] Based on the improved deep Q-network algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information;

[0007] Based on a position matching algorithm that uses orientation and distance, each UAV can smoothly transition from its current formation position to the target sub-formation position.

[0008] Based on an improved multi-group cooperative evolution algorithm, a cooperative trajectory from the starting point to the target point is planned for each sub-formation, and collision detection and trajectory output are performed to complete the UAV formation cooperative mission planning.

[0009] As a preferred method, based on the improved deep Q-network algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information. Specifically, the following steps are included:

[0010] Obtain the initial position, mission target position, and target value of the drone formation, and normalize the target value;

[0011] Calculate the first comprehensive matching degree matrix based on the initial position of the drone formation and the position of the mission target;

[0012] Calculate the number of drones that should be allocated to each mission objective based on the normalized target value;

[0013] An improved deep Q-network is constructed to generate the optimal target allocation scheme based on the number of drones to be allocated to each task objective and the first comprehensive matching degree matrix.

[0014] As a preferred method, the calculation of the first comprehensive matching degree matrix includes the following steps:

[0015] Calculate the initial positional relationship matrix between UAVs and the positional relationship matrix between mission targets;

[0016] Calculate the orientation matching degree between each UAV and each mission target; the calculation rule for the orientation matching degree is: calculate the relative distance matrix between UAVs in the UAV formation. Relative distance matrix between UAV and mission target Among them, the inter-machine relative distance matrix elements in For drones and drones Three-dimensional distances before formation switching, relative distance matrix elements in For drones target location and drones target location The three-dimensional distance, where, This indicates the relative distance between the two drones. , and Indicates drone and drones In horizontal Direction, Horizontal The relative distance components in the direction of height and altitude. , and Indicates drone target location and drones target location In horizontal Direction, Horizontal The relative distance components in the direction of direction and height, with superscripts. Indicates transpose; for the relative distance matrix between drone formations. Relative distance matrix between UAV and mission target Construct distance mapping matrix ,in:

[0017]

[0018] in, Represents the relative distance matrix between drone formations The elements in the matrix or the relative distance matrix from the UAV to the mission target The values ​​of the elements in the array after being processed by the mapping function. This indicates the total number of values ​​after the mapping function has processed them. This indicates the relative distance between two drones or the relative distance between target locations. Represents any numerical value, distance mapping matrix Reflecting the position of the drone within the formation, based on the distance mapping matrix Extract the positional information of the drones within the formation. and the positional information of the mission target relative to the UAV Based on the concept of set matching degree, calculation is performed row by row. right The set matching degree of each row is used to derive the corresponding orientation matching degree matrix. ;

[0019] Calculate the Euclidean distance between the UAV and the mission target, and convert it into a distance score;

[0020] The first comprehensive matching degree matrix is ​​obtained by normalizing and weighting the orientation matching degree and distance score.

[0021] As a preferred embodiment, the structure of the improved deep Q-network includes an input layer, a first fully connected layer, a first batch normalization layer, a first activation function layer, a first dropout layer, a second fully connected layer, a second batch normalization layer, a second activation function layer, a second dropout layer, a third fully connected layer, a third activation function layer, a fourth fully connected layer, a fourth activation function layer, and an output layer connected in sequence.

[0022] As a preferred method, a position matching algorithm based on azimuth and distance is used to smoothly transition each UAV from its current formation position to the target sub-formation formation position, specifically as follows:

[0023] Based on the current sub-formation formation matrix and the preset new target formation matrix, the second comprehensive matching degree matrix is ​​calculated using the calculation method of the first comprehensive matching degree matrix;

[0024] Based on the second comprehensive matching degree matrix, the Hungarian algorithm is applied to solve for the optimal allocation and output the optimal position allocation result.

[0025] As a preferred approach, an improved multi-group co-evolutionary algorithm is used to plan a cooperative trajectory from the starting point to the target point for each sub-formation, specifically as follows:

[0026] Based on digital elevation maps, a three-dimensional terrain model is generated through scale conversion to clarify terrain elevation constraints.

[0027] Random sampling generation in a 3D terrain model 1 sampling point, filter out the collision-free sampling points;

[0028] Calculate the distance between any two points in the collision-free sampling points. If the distance between the two points is less than a preset threshold and the line connecting them is collision-free, then construct a path segment to form the initial PRM network.

[0029] The starting and ending points of each sub-formation are added as new nodes and connected to the initial PRM network. The connection relationship between the new nodes and the nodes of the initial PRM network is calculated, and collision-free path segments are filtered out to complete the PRM network expansion and obtain a complete route map containing all starting and ending points.

[0030] Calculate the angle matrix between the collision-free path segment and the line connecting the "starting point and the target point", and use the reciprocal of the angle as the angle heuristic information for path search to accelerate the search for the optimal path;

[0031] Traverse all path segments in the expanded PRM network, calculate the distance between each sampling point, and construct a distance matrix.

[0032] All path segments in the expanded PRM network are uniformly sampled, and threat calculation is performed on each sample point based on the threat function. Then, the threat values ​​of the sample points are accumulated to construct a sparse threat matrix.

[0033] Based on the extended PRM network, angle heuristic information, distance matrix, initial population of each UAV and sparse threat matrix, calculate the individual fitness function, and execute the individual genetic algorithm on the population of each UAV according to the individual fitness function to obtain the optimal individual of each UAV population, i.e. the trajectory of each UAV.

[0034] Construct a combined solution based on the optimal individual of each UAV swarm, calculate the comprehensive cost function of the combined solution, and execute the elite retention strategy and swarm update based on the comprehensive cost function to obtain the globally optimal combined solution;

[0035] Collision detection is performed on the globally optimal combined solution, and the track that passes all collision detections is output to obtain the global track of the UAV formation.

[0036] Preferably, the three-dimensional terrain model includes the radar detection zone, the first threat zone, the second threat zone, the no-fly zone, and the radar blind zone;

[0037] Radar detection rays The calculation formula is:

[0038]

[0039] in, Indicates the radar station's altitude, determined by the radar's position. Expand outwards sequentially, if Terrain height If the location is below the radar detection ray, it indicates a radar blind zone, with a height value of [value missing]. , If the terrain height If the location is above the radar detection ray, the radar detection ray is updated and continues to expand outward to calculate the blind zone height;

[0040] Threat function in the first threat zone for:

[0041]

[0042] in, The distance between the primary threat source launcher and the target. The maximum attack range of the primary threat source. Minimum attack range for the primary threat source The primary threat source interception coverage area;

[0043] Threat function in the second threat zone for:

[0044]

[0045] in, The coordinates of the second threat source position. The coordinates of the second threat zone. The maximum firing radius of the second threat source. For the effective threat space of the second threat source, This represents the natural index.

[0046] As a preferred option, the formula for calculating the fitness function of a single machine is:

[0047]

[0048] in, Indicates the first The cost of a single drone aircraft. Indicates the planned flight path is It is formed by connecting segments of the flight path. Indicating the first in the planned flight path The length of the flight path, Indicating the first in the planned flight path The cost of the threats received by Duan Hangji Indicates the first The flight constraints of the drone's planned flight path are violated. This represents the penalty coefficient for violations of flight constraints.

[0049] As a preferred approach, a combined solution is constructed, the comprehensive cost function of the combined solution is calculated, and an elite retention strategy and population update are executed based on the comprehensive cost function. This process includes the following steps:

[0050] Based on the individuals in each population, construct a combined solution of "current individual + best individual from other populations";

[0051] Calculate the combined cost function of the combined solution:

[0052]

[0053] in, This represents the combined cost of the solution. Indicates the first The cost of a single drone's flight path. This represents the time-coherence violation of the combined solution. This represents the cooperative violation quantity in the combined solution space. As a penalty factor, Indicates the total number of sub-groups;

[0054] Find the optimal combination solution with the best overall cost after this evolution. Compare the optimal combination solution after this evolution with the global optimal combination solution. If the overall cost of the optimal combination solution after this evolution is less than the overall cost of the global optimal combination solution, then update the global optimal combination solution to the optimal combination solution after the current evolution; otherwise, do nothing.

[0055] The beneficial effects of this invention are:

[0056] 1. This invention constructs an autonomous process for switching from the original formation collaboration to the sub-formation collaboration target through a hierarchical intelligent method, which significantly improves the mission planning efficiency and combat capability of UAV formations in complex scenarios.

[0057] 2. This invention achieves autonomous task planning for UAV formations in saturated scenarios through a hierarchical intelligent approach: In the task allocation phase, the improved DQN algorithm enhances allocation efficiency and target value utilization; in the formation switching phase, a matching algorithm based on orientation and distance ensures a smooth transition; and in the trajectory planning phase, an improved multi-population co-evolutionary algorithm avoids local optima through co-evolution, improving trajectory safety and coordination. The overall approach provides reliable technical support for the application of UAV formations. Attached Figure Description

[0058] Figure 1 This is a flowchart of the UAV formation collaborative task planning method based on hierarchical intelligence according to the present invention.

[0059] Figure 2 This is a control architecture diagram of the hierarchical intelligent algorithm of the present invention.

[0060] Figure 3 The assignment results are based on the improved DQN algorithm.

[0061] Figure 4 This is a top view of the trajectory planning based on an improved multi-population co-evolutionary algorithm.

[0062] Figure 5 A 3D view of the trajectory planning based on an improved multi-population co-evolutionary algorithm.

[0063] Figure 6 The overall cost curve for the improved multi-population co-evolutionary algorithm. Detailed Implementation

[0064] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0065] Example 1:

[0066] like Figure 1 As shown, a hierarchical intelligence-based UAV formation cooperative task planning method includes the following steps:

[0067] S1. Based on the improved Deep Q-Network (DQN) algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information;

[0068] S2. A position matching algorithm based on azimuth and distance smoothly transitions each UAV from its current formation position to the target sub-formation position;

[0069] S3. Based on an improved multi-group cooperative evolution algorithm, a cooperative trajectory from the starting point to the target point is planned for each sub-formation, and collision detection and trajectory output are performed to complete the UAV formation cooperative mission planning.

[0070] In this embodiment, based on the improved deep Q-network algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information. Specifically, the steps include:

[0071] S11. Parameter Setting Steps. Set the algorithm parameters and reward weights. Algorithm parameters include learning rate, learning rate decay, exploration rate, exploration rate decay, number of training rounds, and batch training samples, etc.; reward weights include coverage weight, matching degree weight, and target value reward weight, etc.

[0072] S12. Data Acquisition Steps. Acquire the initial position of the drone formation, the position of the mission target, and the target value, and normalize the target value.

[0073] S13. Calculation steps for the azimuth matching matrix. Based on the initial position of the UAV and the position of the mission target, calculate the comprehensive azimuth matching matrix, which includes the following steps:

[0074] S131. Calculate the initial positional relationship matrix between UAVs and the positional relationship matrix between mission targets;

[0075] S132. Calculate the orientation matching degree between each UAV and each mission target; the calculation rule for the orientation matching degree is: calculate the relative distance matrix between UAVs in the UAV formation. Relative distance matrix between UAV and mission target Among them, the inter-machine relative distance matrix elements in For drones and drones Three-dimensional distances before formation switching, relative distance matrix elements in For drones target location and drones target location The three-dimensional distance, where, This indicates the relative distance between the two drones. , and Indicates drone and drones In horizontal Direction, Horizontal The relative distance components in the direction of height and altitude. , and Indicates drone target location and drones target location In horizontal Direction, Horizontal The relative distance components in the direction of direction and height, with superscripts. Indicates transpose; for the relative distance matrix between drone formations. Relative distance matrix between UAV and mission target Construct distance mapping matrix ,in:

[0076]

[0077] in, Represents the relative distance matrix between drone formations The elements in the matrix or the relative distance matrix from the UAV to the mission target The values ​​of the elements in the array after being processed by the mapping function. This indicates the total number of values ​​after the mapping function has processed them. This indicates the relative distance between two drones or the relative distance between target locations. Represents any numerical value, distance mapping matrix Reflecting the position of the drone within the formation, based on the distance mapping matrix Extract the positional information of the drones within the formation. and the positional information of the mission target relative to the UAV Based on the concept of set matching degree, calculation is performed row by row. right The set matching degree of each row is used to derive the corresponding orientation matching degree matrix. ;

[0078] S133. Calculate the Euclidean distance between the UAV and the mission target, and convert it into a distance score;

[0079] S134. Normalize and weight the orientation matching degree and distance score to obtain the first comprehensive matching degree matrix.

[0080] S14. Ideal Allocation Calculation Steps. Based on the normalized target value, calculate the number of drones that should be allocated to each task objective, specifically including the following steps:

[0081] S141. Calculate the total value;

[0082] S142. Calculate the number of drones to be allocated based on the proportion of the value of each mission objective to the total value, and ensure that the total number equals the number of drones.

[0083] S15. Improved Deep Q-Network Initialization Steps: Constructing a Deep Neural Network. The structure of the improved Deep Q-Network includes, in sequence, an input layer, a first fully connected layer, a first batch of normalized layers, a first activation function layer, a first dropout layer, a second fully connected layer, a second batch of normalized layers, a second activation function layer, a second dropout layer, a third fully connected layer, a third activation function layer, a fourth fully connected layer, a fourth activation function layer, and an output layer.

[0084] S16. Experience replay pool initialization steps;

[0085] S17. Network Training Steps. Initialize the state using a high-value-first strategy, select actions using a balanced action selection strategy, execute actions to obtain the next state, calculate the reward using an improved reward function, store the experience, update the network using a high-value-first experience replay strategy, periodically update the target network, and dynamically adjust the parameters using an adaptive parameter adjustment strategy.

[0086] The S17 high-value-first experience replay strategy is specifically based on experience coverage. Orientation matching degree and high-value scores The sampling weights are dynamically adjusted, prioritizing coverage and location matching in the early stages, and focusing on high-value allocation in the later stages. Sampling weights The calculation formula is:

[0087]

[0088] Among them, the weighting coefficient Adjustments are made dynamically based on the training phase. Early stage: Later stages: In the early stages, the focus is on coverage and location matching; in the later stages, the focus is on high-value allocation.

[0089] The adaptive parameter adjustment strategy in S17 is as follows: monitor the performance metrics of the most recent rounds, and if the coverage, orientation matching degree or high value score is lower than the threshold, increase the corresponding reward weight; if the reward stagnates, increase the exploration rate and periodically decay the learning rate.

[0090] S18. Optimal strategy generation step: Generate the optimal task allocation scheme through the trained network.

[0091] In this embodiment, a position matching algorithm based on azimuth and distance is used to smoothly transition each UAV from its current formation position to the target sub-formation formation position, including the following steps:

[0092] S21. Input the current sub-formation formation matrix and the new target formation matrix, and use the same orientation matching degree matrix calculation method as S13 to obtain the second comprehensive matching degree matrix.

[0093] S22. Based on the second comprehensive matching degree matrix, apply the Hungarian algorithm to solve for the optimal allocation and output the optimal position allocation result.

[0094] In this embodiment, step S3, which involves coordinating the flight paths of multiple sub-formations to the target point based on an improved multi-population co-evolutionary algorithm, includes the following sub-steps:

[0095] S31. 3D Environment Modeling. Based on a digital elevation map, a 3D terrain model is generated through scale conversion, clearly defining terrain elevation constraints; the radar detection zone, primary threat zone, secondary threat zone, and no-fly zone are defined, and the radar blind zone is calculated. Among these, the radar detection ray... The calculation formula is:

[0096]

[0097] in, Indicates the radar station's altitude, determined by the radar's position. Expand outwards sequentially, if Terrain height If the location is below the radar detection ray, it indicates a radar blind zone, with a height value of [value missing]. , If the terrain height If located above the radar detection ray, the radar detection ray is updated and continues to expand outwards to calculate the blind zone height; if within The terrain elevation is below the radar detection ray, creating a blind zone. The blind zone height is [value missing]. ,Right now height below the radar detection ray minus At the terrain elevation, express At the terrain elevation, express elevation at the location;

[0098] Threat function in the first threat zone for:

[0099]

[0100] in, The distance between the primary threat source launcher and the target. The maximum attack range of the primary threat source. Minimum attack range for the primary threat source The primary threat source interception coverage area;

[0101] Threat function in the second threat zone for:

[0102]

[0103] in, The coordinates of the second threat source position. The coordinates of the second threat zone. The maximum firing radius of the second threat source. This constitutes an effective threat space for the second source of threat.

[0104] S32. PRM Probabilistic Roadmap Construction and Start / End Point Integration. Specifically, it involves generating a probabilistic roadmap in a 3D environment using random sampling. Each sampling point is selected through collision detection to identify collision-free sampling points. For the selected sampling points, if the distance between two points is less than a set threshold and the connection line is collision-free, a path segment is constructed to form the initial PRM network. The starting point and target point of each sub-group are added as new nodes to the PRM network. The connection relationship between the new nodes and existing network nodes is calculated, collision-free path segments are selected, and the PRM network is expanded to obtain a complete route map containing all start and end points.

[0105] S33. Calculate the angle heuristic, distance matrix, and threat matrix. Specifically:

[0106] Angle heuristic information: Calculate the angle matrix between the path segment and the line connecting the starting point and the target point. Use the reciprocal of the angle as a heuristic factor for path search to accelerate the search for the optimal path. The formula for calculating angle heuristic information is:

[0107]

[0108] in, Indicates starting from the current node Select neighboring nodes The probability of that, i.e., the angle-heuristic information. Indicates the current node and different neighboring nodes The angle between the connecting line segment and the baseline, where the baseline is the straight line from the starting point to the target point. For nodes All neighboring nodes. From the above equation, we can see that neighboring nodes... The probability of being selected and Inversely proportional, The smaller the value, the more likely it is to select a neighboring node. The greater the likelihood, the higher the probability.

[0109] Distance Matrix and Threat Matrix: Traverse all path segments in the PRM network, calculate the distance between each sampling point, and construct the distance matrix; uniformly sample each path segment, and calculate the threat for each sampling point based on the threat function, then accumulate the threat values ​​of the sampling points to finally construct a sparse threat matrix. In the sparse threat matrix, if there is no threat, it is 0; if there is a threat, the threat value is calculated based on the threat type and its function corresponding to the sampling point, providing data for fitness calculation in subsequent path planning algorithms;

[0110] Based on the extended PRM network, angle heuristic information, distance matrix, initial population of each UAV and sparse threat matrix, calculate the individual fitness function, and execute the individual genetic algorithm on the population of each UAV according to the individual fitness function to obtain the optimal individual of each UAV population, i.e. the trajectory of each UAV.

[0111] S34. Based on the extended PRM network, angle heuristic information, distance matrix, and the initial population of each UAV, calculate the individual fitness function, and execute the individual genetic algorithm on the population of each UAV according to the individual fitness function to obtain the optimal individual of each UAV population, i.e., the trajectory of each UAV, including the following sub-steps:

[0112] Population initialization: Based on the PRM network and angle heuristic information, a direction heuristic factor and a target point reverse search strategy are introduced to generate the initial population. The target point reverse search first searches the neighboring nodes of the target node to form the target node's neighboring node set. During the track search, if the search reaches the range of the target node's neighboring node set, a node is selected from it and connected to the target node, thereby completing the track search.

[0113] Calculate the fitness function for a single machine:

[0114]

[0115] in, Indicates the first The cost of a single drone aircraft. Indicates the planned flight path is It is formed by connecting segments of the flight path. Indicating the first in the planned flight path The length of the flight path, Indicating the first in the planned flight path The cost of the threats received by Duan Hangji Indicates the first The flight constraints of the drone's planned flight path are violated. This represents the penalty coefficient for violations of flight constraints.

[0116] Single-machine genetic algorithm operation: Use the two-race method to select high-quality individuals while retaining the best individuals of each population; randomly select the common node of two individuals as the crossover point, swap the path segments after the crossover point, and delete the looping nodes in the path; force mutation on the high-quality individuals, randomly select two mutation points in the path, and reconnect the mutation points with the start and end points through the shortest path search to generate a new path.

[0117] S35. Construct a combined solution based on the optimal individuals of each UAV swarm, calculate the comprehensive cost function of the combined solution, and execute the elite preservation strategy and swarm update based on the comprehensive cost function to obtain the globally optimal combined solution; specifically including:

[0118] Constructing a combined solution: For each individual in the population, construct a combined solution of "the current individual + the best individual in other populations".

[0119] Calculate the combined cost function of the defined solution:

[0120]

[0121] in, This represents the combined cost of the solution. Indicates the first The cost of a single drone's flight path. This represents the time-coherence violation of the combined solution. This represents the cooperative violation quantity in the combined solution space. As a penalty factor, This indicates the total number of sub-teams.

[0122] Elite Preservation and Population Update: Find the optimal combination solution with the best overall cost after this evolution; compare the optimal combination solution after this evolution with the global optimal combination solution; if the overall cost of the optimal combination solution after this evolution is less than the overall cost of the global optimal combination solution, then update the global optimal combination solution to the optimal combination solution after the current evolution, otherwise do not process it.

[0123] S36. Perform collision detection on the globally optimal combination solution, output the trajectory that passes all collision detections, and obtain the global trajectory of the UAV formation.

[0124] like Figure 2 As shown, this invention realizes a hierarchical intelligent UAV formation cooperative task planning method based on an improved DQN algorithm at the task allocation layer and an improved multi-group cooperative evolution algorithm at the trajectory planning layer.

[0125] Define a target area containing four high-value targets with values ​​of 100, 90, 80, and 70 respectively. Deploy a triangular formation of 16 drones and run an improved DQN allocation algorithm to verify its allocation performance. Figure 3 It can be seen that by applying the improved DQN algorithm, one drone is assigned to the highest-value target 1, four drones to the second-highest-value target 2, four drones to target 3, and three drones to the lowest-value target 4. This achieves a high-value priority scheme with full target coverage and an average azimuth matching degree of 0.897. The drone formation and target azimuth matching degree is relatively high, providing a safer foundation for formation switching and subsequent trajectory planning. Figure 4 and Figure 5 The flight path planning curves of the four sub-formations to the target are shown. All flight paths successfully avoided high-threat areas and no-fly zones. Figure 6The comprehensive cost curve is presented, which integrates track length, threat cost, flight constraint violation, and spatiotemporal cooperation cost among formations. In the early stages of evolution, thanks to the directional heuristic factor and a high-quality initial population, the cost decreases rapidly. In the middle stages of evolution, through multi-population co-evolution and elite retention strategies, the algorithm continuously explores better solutions, and the cost steadily decreases. In the later stages of evolution, the algorithm tends to converge, and the comprehensive cost stabilizes at a low level. This demonstrates that the improved multi-population co-evolutionary track planning algorithm proposed in this invention can effectively escape local optima and efficiently plan UAV formation tracks with excellent overall performance and meeting cooperation requirements.

[0126] Example 2:

[0127] Based on Embodiment 1, this embodiment of the invention provides a hierarchical intelligence-based UAV formation cooperative mission planning system, which can be used to implement the hierarchical intelligence-based UAV formation cooperative mission planning method as described in the foregoing embodiments. The system includes:

[0128] The first module is used to split the original formation into multiple sub-formations and assign targets based on the improved deep Q-network algorithm according to formation characteristics and target value information;

[0129] The second module is used for a position matching algorithm based on azimuth and distance to smoothly transition each UAV from its current formation position to the target sub-formation position.

[0130] The third module is used to plan the cooperative trajectory from the starting point to the target point for each sub-formation based on the improved multi-group cooperative evolution algorithm, and to perform collision detection and trajectory output to complete the UAV formation cooperative mission planning.

[0131] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0132] In an exemplary embodiment, the electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the hierarchical intelligence-based UAV formation cooperative task planning method as described in Embodiment 1 above.

[0133] In an exemplary embodiment, the readable storage medium may be a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the hierarchical intelligence-based UAV formation cooperative task planning method according to Embodiment 1 above.

[0134] In an exemplary embodiment, the computer program product includes a computer program that, when executed by a processor, implements the hierarchical intelligence-based UAV formation cooperative task planning method described in Embodiment 1 above.

[0135] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0136] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0137] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0138] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0139] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0140] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for collaborative task planning in UAV formation based on hierarchical intelligence, characterized in that, Includes the following steps: Based on the improved deep Q-network algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information; Based on a position matching algorithm that uses orientation and distance, each UAV can smoothly transition from its current formation position to the target sub-formation position. Based on an improved multi-group cooperative evolution algorithm, a cooperative trajectory from the starting point to the target point is planned for each sub-formation, and collision detection and trajectory output are performed to complete the UAV formation cooperative mission planning. Based on an improved multi-swarm cooperative evolution algorithm, a cooperative trajectory from the starting point to the target point is planned for each sub-formation, specifically as follows: Based on digital elevation maps, a three-dimensional terrain model is generated through scale conversion to clarify terrain elevation constraints. Random sampling generation in a 3D terrain model 1 sampling point, filter out the collision-free sampling points; Calculate the distance between any two points in the collision-free sampling points. If the distance between the two points is less than a preset threshold and the line connecting them is collision-free, then construct a path segment to form the initial PRM network. The starting and ending points of each sub-formation are added as new nodes and connected to the initial PRM network. The connection relationship between the new nodes and the nodes of the initial PRM network is calculated, and collision-free path segments are filtered out to complete the PRM network expansion and obtain a complete route map containing all starting and ending points. Calculate the angle matrix between the collision-free path segment and the line connecting the "starting point and the target point", and use the reciprocal of the angle as the angle heuristic information for path search to accelerate the search for the optimal path; Traverse all path segments in the expanded PRM network, calculate the distance between each sampling point, and construct a distance matrix. All path segments in the expanded PRM network are uniformly sampled, and threat calculation is performed on each sample point based on the threat function. Then, the threat values ​​of the sample points are accumulated to construct a sparse threat matrix. Based on the extended PRM network, angle heuristic information, distance matrix, initial population of each UAV and sparse threat matrix, calculate the individual fitness function, and execute the individual genetic algorithm on the population of each UAV according to the individual fitness function to obtain the optimal individual of each UAV population, i.e. the trajectory of each UAV. Construct a combined solution based on the optimal individual of each UAV swarm, calculate the comprehensive cost function of the combined solution, and execute the elite retention strategy and swarm update based on the comprehensive cost function to obtain the globally optimal combined solution; Collision detection is performed on the globally optimal combined solution, and the track that passes all collision detections is output to obtain the global track of the UAV formation.

2. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 1, characterized in that, Based on the improved deep Q-network algorithm, the original formation is split into multiple sub-formations and targets are assigned according to formation characteristics and target value information. The specific steps include: Obtain the initial position, mission target position, and target value of the drone formation, and normalize the target value; Calculate the first comprehensive matching degree matrix based on the initial position of the drone formation and the position of the mission target; Calculate the number of drones that should be allocated to each mission objective based on the normalized target value; An improved deep Q-network is constructed to generate the optimal target allocation scheme based on the number of drones to be allocated to each task objective and the first comprehensive matching degree matrix.

3. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 2, characterized in that, The calculation of the first comprehensive matching degree matrix includes the following steps: Calculate the initial positional relationship matrix between UAVs and the positional relationship matrix between mission targets; Calculate the orientation matching degree between each UAV and each mission target; the calculation rule for the orientation matching degree is: calculate the relative distance matrix between UAVs in the UAV formation. Relative distance matrix between UAV and mission target Among them, the inter-machine relative distance matrix elements in For drones and drones Three-dimensional distances before formation switching, relative distance matrix elements in For drones target location and drones target location The three-dimensional distance, where, This indicates the relative distance between the two drones. , and Indicates drone and drones In horizontal Direction, Horizontal Direction, height The relative distance component in the direction, , and Indicates drone target location and drones target location In horizontal Direction, Horizontal Direction, height The relative distance component in the direction, superscript Indicates transpose; for the relative distance matrix between drone formations. Relative distance matrix between UAV and mission target Construct distance mapping matrix ,in: in, Represents the relative distance matrix between drone formations The elements in the matrix or the relative distance matrix from the UAV to the mission target The values ​​of the elements in the array after being processed by the mapping function. This indicates the total number of values ​​after the mapping function has processed them. This indicates the relative distance between two drones or the relative distance between target locations. Represents any numerical value, distance mapping matrix Reflecting the position of the drone within the formation, based on the distance mapping matrix Extract the positional information of the drones within the formation. and the positional information of the mission target relative to the UAV Based on the concept of set matching degree, calculation is performed row by row. right The set matching degree of each row is used to derive the corresponding orientation matching degree matrix. ; Calculate the Euclidean distance between the UAV and the mission target, and convert it into a distance score; The first comprehensive matching degree matrix is ​​obtained by normalizing and weighting the orientation matching degree and distance score.

4. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 2, characterized in that, The improved deep Q-network structure includes an input layer, a first fully connected layer, a first batch normalization layer, a first activation function layer, a first dropout layer, a second fully connected layer, a second batch normalization layer, a second activation function layer, a second dropout layer, a third fully connected layer, a third activation function layer, a fourth fully connected layer, a fourth activation function layer, and an output layer, connected in sequence.

5. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 2, characterized in that, Based on a position matching algorithm that uses azimuth and distance, each UAV smoothly transitions from its current formation position to the target sub-formation position, specifically as follows: Based on the current sub-formation formation matrix and the preset new target formation matrix, the second comprehensive matching degree matrix is ​​calculated using the calculation method of the first comprehensive matching degree matrix; Based on the second comprehensive matching degree matrix, the Hungarian algorithm is applied to solve for the optimal allocation and output the optimal position allocation result.

6. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 1, characterized in that, The three-dimensional terrain model includes the radar detection zone, the first threat zone, the second threat zone, the no-fly zone, and the radar blind zone; Radar detection rays The calculation formula is: in, Indicates the radar station's altitude, determined by the radar's position. Expand outwards sequentially, if Terrain height If the location is below the radar detection ray, it indicates a radar blind zone, with a height value of [value missing]. , If the terrain height If the location is above the radar detection ray, the radar detection ray is updated and continues to expand outward to calculate the blind zone height; Threat function in the first threat zone for: in, The distance between the primary threat source launcher and the target. The maximum attack range of the primary threat source. Minimum attack range for the primary threat source The primary threat source interception coverage area; Threat function in the second threat zone for: in, The coordinates of the second threat source position. The coordinates of the second threat zone. The maximum firing radius of the second threat source. For the effective threat space of the second threat source, This represents the natural index.

7. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 1, characterized in that, The formula for calculating the fitness function of a single machine is: in, Indicates the first The cost of a single drone aircraft. Indicates the planned flight path is It is formed by connecting segments of the flight path. Indicates the first in the planned flight path The length of the flight path, Indicating the first in the planned flight path The cost of the threats received by Duan Hangji Indicates the first The flight constraints of the drone's planned flight path are violated. This represents the penalty coefficient for violations of flight constraints.

8. The UAV formation cooperative task planning method based on hierarchical intelligence according to claim 1, characterized in that, Constructing a combined solution, calculating the comprehensive cost function of the combined solution, and implementing an elite retention strategy and population update based on the comprehensive cost function, specifically includes the following steps: Based on the individuals in each population, construct a combined solution of "current individual + best individual from other populations"; Calculate the combined cost function of the combined solution: in, This represents the combined cost of the solution. Indicates the first The cost of a single drone's flight path. This represents the time-coherence violation of the combined solution. This represents the cooperative violation quantity in the combined solution space. As a penalty factor, Indicates the total number of sub-groups; Find the optimal combination solution with the best overall cost after this evolution. Compare the optimal combination solution after this evolution with the global optimal combination solution. If the overall cost of the optimal combination solution after this evolution is less than the overall cost of the global optimal combination solution, then update the global optimal combination solution to the optimal combination solution after the current evolution; otherwise, do nothing.

Images