Double-layer and double-stage heterogeneous unmanned aerial vehicle task allocation and flight path planning method

Abstract

The invention discloses a double-layer and double-stage heterogeneous unmanned aerial vehicle task allocation and flight path planning method, and relates to the technical field of unmanned aerial vehicles. The method comprises a dual-stage task allocation method and a dual-stage path planning method. The beneficial effects of the invention are that the dual-stage task allocation method and the dual-stage path planning method are provided for improving the efficiency of the multi-unmanned aerial vehicle cooperative execution of the search rescue task; according to the dual-stage task allocation method, a joint optimization framework combining mixed integer linear programming and an improved ant colony algorithm is provided, so that the task load balance and the total flight distance can be optimized, and a better task allocation effect is realized; the dual-stage path planning method provides a dual-stage path planning scheme in which global task sequence optimization and local obstacle avoidance planning are coordinated, and a better path planning effect is obtained by combining the global and local dual-stage optimization scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicles, in particular to a double-layer double-stage heterogeneous unmanned aerial vehicle task allocation and path planning method. BACKGROUND

[0002] Multi-unmanned aerial vehicle task allocation aims to efficiently match limited unmanned aerial vehicle resources with task demands in dynamic environments, with the core goal of optimizing task completion efficiency, resource utilization, and system robustness. Existing technologies mainly fall into the following two categories: (1) Centralized task allocation: 1) Mixed integer linear programming (MILP): By constructing a mathematical model (such as minimizing total flight distance as the objective function, with constraints including task integrity, payload limitations, etc.), the branch and bound method is used to quickly solve the global optimal task allocation scheme. The timing coordination and resource coordination constraints in unmanned aerial vehicle formation cooperative combat.

[0003] 2) Clustering algorithm: According to the geographical distribution or attribute similarity of task points, task subsets are divided to reduce the problem dimension, which is suitable for regional coverage tasks.

[0004] 3) Intelligent optimization algorithm: including genetic algorithm, particle swarm algorithm, ant colony algorithm, etc. Some scholars have proposed an improved route planning algorithm based on genetic algorithm, which optimizes the chromosome coding by introducing the K-means algorithm to improve the target search capability; some scholars have improved the particle swarm algorithm, using a segmented inertia weight adjustment strategy to enhance the multi-objective path planning capability.

[0005] (2) Distributed task allocation Game theory method: Through a non-cooperative game model, the dynamic balance of task benefits is achieved, but there is a risk of strategy space dimension explosion.

[0006] Contract net protocol (CNP): adopts a three-stage mechanism of bidding-tender-bidding, supporting heterogeneous unmanned aerial vehicle task coordination, but is prone to efficiency decay due to communication load overload.

[0007] Auction algorithm: draws on the competitive mechanism to achieve optimal task matching, with fast response speed in time-sensitive tasks, but requires complex benefit allocation rules.

[0008] 2. Current status of unmanned aerial vehicle path planning technology Unmanned aerial vehicle path planning aims to generate the optimal flight path from the starting point to the task point under the constraints of safety, load, etc. Existing technologies mainly include: (1) Linear programming and graph search algorithm Mixed integer linear optimization (MILO): a 0-1 MILO model based on geometric transformation is used for airspace collision avoidance, with fast processing speed and real-time requirements.

[0009] Dijkstra's algorithm: Achieve shortest path search by expanding nodes through priority queue, but with high computational complexity, difficult to adapt to dynamic environment.

[0010] A* algorithm: Introduce heuristic function to improve search efficiency, but prone to local optimum in high-dimensional space.

[0011] D* algorithm: Support local re-planning in dynamic environment, but only modify nodes affected by environmental changes.

[0012] (2) Intelligent optimization and sampling algorithm Rapidly-exploring Random Tree (RRT): Explore high-dimensional space through probabilistic sampling and tree expansion, with dimension-independent and probabilistic completeness. Although RRT algorithm is used to plan conflict-free path, the path has many redundant turning points and poor smoothness.

[0013] Ant Colony Algorithm: Based on pheromone positive feedback mechanism to optimize path, suitable for multi-objective planning, but slow initial convergence speed and large computational resource consumption. Introduce deviation distance factor to optimize heuristic function, improve algorithm's ability to jump out of local optimum.

[0014] Genetic Algorithm: Simulate biological evolution mechanism, approximate global optimum through chromosome coding and fitness function, but parameter setting directly affects convergence performance.

[0015] (3) Reinforcement learning algorithm Q-Learning: Based on dynamic programming to realize policy iteration, without prior environmental model, but weak generalization ability for high-dimensional state space.

[0016] Deep Reinforcement Learning (DRL): Such as Deep Q Network (DQN), combine deep learning to handle complex environmental state, but high training cost and rely on large amount of data.

[0017] Although existing technologies have made some progress in multi-UAV task allocation and path planning, there are still the following core problems: (1) Task allocation efficiency and load balancing are insufficient MILP model limitations: Traditional MILP method only solves global task allocation of multi-start unmanned aerial vehicles, does not consider task load balancing of unmanned aerial vehicles with the same starting point, and the task load standard deviation can reach 2.2750 under MILP single-stage allocation, leading to some unmanned aerial vehicles overloading, affecting task execution continuity.

[0018] Intelligent algorithm convergence defects: Ant colony algorithm has slow initial convergence speed, genetic algorithm is prone to local optimum, and lacks dynamic adaptability optimization for multi-UAV collaborative scenarios.

[0019] (2) High path planning redundancy and insufficient motion constraints Defects of RRT algorithm: The traditional RRT generates a path with a large number of redundant turning points, the path length is relatively long, and the minimum turning radius constraint of the unmanned aerial vehicle is not considered, resulting in poor path smoothness.

[0020] Lack of global path optimization: Most methods only focus on local obstacle avoidance and do not combine global task sequence optimization (such as the traveling salesman problem TSP), resulting in path planning results deviating from the optimal solution.

[0021] (3) Poor adaptability to dynamic environment Static environment assumption limitation: Most existing technologies are based on static obstacle scenarios, lack real-time re-planning capability for dynamic obstacles (such as moving targets and temporary no-fly zones), and cannot meet the demand for sudden changes in rescue environment.

[0022] Insufficient multi-objective trade-off: Most methods optimize in the direction of a single objective (such as shortest path), without considering the coordinated optimization of time cost, energy consumption, rescue effect, etc. SUMMARY

[0023] The purpose of the present application is to overcome the deficiencies in the prior art, and to provide a dual-layer dual-stage heterogeneous unmanned aerial vehicle task allocation and path planning method. The present application aims to improve the efficiency of multiple unmanned aerial vehicles in executing search and rescue tasks cooperatively, and provides a dual-stage task allocation method and a dual-stage path planning method. The dual-stage task allocation method proposes a joint optimization framework combining mixed integer linear programming and an improved ant colony algorithm to optimize task load balancing and total flight distance, achieving better task allocation results. The dual-stage path planning method proposes a dual-stage path planning scheme that combines global task sequence optimization and local obstacle avoidance planning. By combining global and local dual-stage optimization schemes, better path planning results are achieved.

[0024] To achieve the above objectives, the present invention adopts the following technical solution: a two-layer, two-stage heterogeneous UAV task allocation and trajectory planning method, comprising a two-stage task allocation method and a two-stage path planning method: The two-stage task allocation method (TADP) proposes a joint optimization framework combining mixed integer linear programming (MILP) and an improved ant colony algorithm. First, MILP modeling is performed to address the characteristics of multi-UAV combat missions, and the model is solved to achieve initial task allocation. Considering that in reality, multiple UAVs often start from the same point, a single-warehouse multiple traveling salesman problem (SD-MTSP) model is constructed for UAV swarms starting from the same point, and an elite strategy is integrated. An improved ant colony algorithm based on 2-opt local optimization design is used for secondary task allocation, optimizing task load balancing and total flight distance to achieve better task allocation results. A two-stage path planning method (PPDP) is proposed: a two-stage path planning scheme that combines global task order optimization and local obstacle avoidance planning. The first stage uses the ant colony algorithm to solve the Traveling Salesman Problem (TSP) to optimize the global access order of UAVs to task points. The second stage combines an improved Rapid Random Tree (RRT) algorithm, introducing target bias strategy and path pruning optimization to generate obstacle avoidance paths that balance safety and economy. By combining global and local two-stage optimization, a better path planning effect is achieved.

[0025] Furthermore, the two-stage task allocation method (TADP) is specifically divided into two core stages: the first stage is global task allocation using mixed-integer linear programming: global task allocation is quickly completed based on a mixed-integer linear programming model, optimizing the total flight distance and task integrity; the second stage is local optimization using an improved ant colony algorithm: a single warehouse multiple traveling salesman problem (SD-MTSP) model is constructed for drone swarms starting from the same point, and the task load balance and total flight distance are further optimized by introducing an elite strategy, 2-opt local optimization, and a dynamic load balancing mechanism.

[0026] Furthermore, the global task allocation for the mixed-integer linear programming is specifically as follows: Problem modeling: Task allocation in the context of multiple drones performing search and rescue missions can typically be described as: drone swarm From each starting point Multiple destinations need to be reached. Execute the task Meanwhile, the following constraints must be met: (1) Task integrity: All tasks to be executed must be assigned and assigned only once; (2) Load constraint: The number of tasks assigned to a single UAV does not exceed its maximum load capacity; (3) Distance optimization: Minimize the total flight distance, considering minimizing the Euclidean distance from the UAV to the task point; Assume that the UAV formation includes Homogeneous drones Set of tasks to be executed ,in Indicates drone initial coordinates, Indicates task Location coordinates, The maximum single-machine workload; decision variables Defined as: To minimize the total flight distance, the objective function is constructed as follows: Among them, distance coefficient The calculation is as follows: Based on the actual requirements of multi-UAV task allocation, the following constraints are established: (1) Task allocation integrity constraint: This constraint ensures that each task (2) The payload capacity of the drone is constrained to be assigned to only one drone. This constraint ensures that the number of tasks assigned to each drone does not exceed its maximum payload. (3) Variable type constraint: From a feasibility perspective, to ensure the existence of a feasible solution for the system, the total payload capacity of the UAV must cover all mission loads, i.e. From the perspective of optimality guarantee, when the solver returns the globally optimal solution, the resulting allocation scheme reaches the theoretical optimum under given constraints. In practical engineering, an approximate optimal solution can be obtained by setting a time limit. Model solution algorithm: Input parameter preprocessing, UAV initial coordinate matrix: Task point coordinate matrix: Distance matrix calculation: Construction of constraint matrices, let the equality constraint matrix be... Its elements satisfy: Among them, the index of the decision variable column Each row corresponds to one task, ensuring that the task is assigned only once; let the inequality constraint matrix... , Each row corresponds to one drone, constraining its total number of missions; the optimal solution transforms the problem into a standard mixed-integer linear programming (MILP) form: ,in , , The flattened distance vector The Highs solver is used for the branch and bound algorithm, and a convergence tolerance is set. To ensure calculation accuracy.

[0027] Furthermore, the improved ant colony algorithm's local optimization specifically involves: Problem modeling: After the initial allocation is completed in the mixed-integer programming, for the same starting point... The problem of secondary task allocation for drones can be modeled as a single-warehouse multiple traveling salesman problem (SD-MTSP). Given a shared initial location... The drone formation, including Homogeneous drones Set of tasks to be executed coordinate set The constraints are: (1) Each task point is visited only once; (2) All UAVs return to their initial positions; (3) The total flight distance is minimized. This problem can be formalized as: ,in Indicates the first The closed-loop path of the drone. express , The Euclidean distance between them; Algorithm design: Based on the traditional ant colony algorithm, and based on the characteristics of SD-MTSP, an improved ant colony algorithm is designed, which includes the following core components: dynamic pheromone update mechanism, adopting an elitist strategy, only the globally optimal path participates in pheromone enhancement, and defining edges. pheromone concentration Update rules: ,in, The pheromone evaporation coefficient; the pheromone concentration increment. , Let pheromone intensity be constant. The shortest path in this round The total path length of a single ant; the path selection probability model, the ant at the node Select the next node The probability of: ,in, As a heuristic factor; Controlling the influence weight of pheromones Control heuristic information weights; load balancing strategy, to avoid uneven task distribution, introduce dynamic load constraints: ,in For the set threshold, The total number of tasks. For the first The number of tasks assigned to each drone is dynamically adjusted during the ant colony search process. The 2-opt algorithm is used for local optimization. The 2-opt algorithm is a local search algorithm for optimizing paths. It improves the quality of the path by swapping two nodes in the path. The ant colony algorithm acts as a global planner to build diverse solutions. The 2-opt algorithm acts as a local planner. In the local search phase, it enhances the solution by refining the path and eliminating inefficient sub-paths. Through the cooperation of global and local algorithms, the task allocation results are optimized.

[0028] Furthermore, the proposed two-stage path planning method (PPDP) specifically addresses multi-UAV collaborative search and rescue missions. Therefore, a PPDP (Path Planning with Dual-Phase Optimization) scheme is proposed. This method achieves an organic combination of global task sequence optimization and local collision avoidance through two-stage collaborative optimization. In the first stage, an ant colony algorithm handles the Task Search Problem (TSP): The ant colony algorithm is used to solve the TSP problem, optimizing the UAV's access order to each task point. In the second stage, an obstacle avoidance algorithm based on an improved fast random tree (RRT) is used: An improved RRT algorithm is employed to generate a safe path that considers obstacles, ensuring that the UAV completes its mission while avoiding obstacles. This two-stage scheme ensures that the UAV path length is as short as possible while significantly improving the smoothness and safety of path planning.

[0029] Furthermore, the ant colony algorithm for handling the Traveling Salesman Problem (TSP) is as follows: After the multi-drone task allocation is completed, each drone has a clear set of task points, thus greatly reducing the complexity of path planning. At this point, the multi-drone path planning problem can be equivalent to multiple independent Traveling Salesman Problems (TSPs), with each TSP instance corresponding to all the task points a drone needs to visit. For this type of path optimization problem, the Ant Colony Optimization (ACO) algorithm is widely used in path planning due to its strong global search capability and adaptability. In the ant colony algorithm, each ant starts from a node and selects the next node to visit based on probability rules until it has visited all nodes and returns to the starting point. The quality of the path is used to guide the updating of pheromones: the shorter the path, the higher the pheromone concentration distributed along it, thus attracting subsequent ants to choose it first. Problem description and model construction: Let the task point set of a certain drone be... ,in This represents the starting point of the drone, and the rest are the assigned task points. Define the distance matrix. ,in Represents a node With nodes Euclidean distance between them: ,in For a minimum value (e.g.) To avoid the abnormal situation of division by zero; define the pheromone matrix. heuristic function Pheromones influence weighting factors Heuristic information weighting factor pheromone volatility coefficient pheromone intensity constant The algorithm involves pheromone concentration increment and dynamic pheromone update mechanism. The solution algorithm is as follows: In each iteration, all ants complete the entire path once, calculate and record the path length, and update the optimal path and global pheromone levels. To improve convergence speed, an elite strategy is introduced: the optimal path (e.g., the globally optimal or iteratively optimal ant) ​​is given a higher pheromone increment. See Algorithm 1 for the specific algorithm details.

[0030] Furthermore, Algorithm 1 specifically refers to: Algorithm 1: Ant Colony Algorithm for TSP Problem; Input: , , , , , UAV mission point coordinates set Output: Optimal path node sequence Initialization: by formula Calculate the distance matrix Pheromone matrix, pheromone matrix Heuristic functions Global optimal path ,length ;1:for to do;2: foreach ants to do;3: Randomly select the starting node and construct the initial tabu list. ;4:for to do;5: From formula Calculate the current node To the set of all unvisited nodes 6: Randomly select the next node according to the probability distribution and add it to the transition probability; ;7: endfor8: calculate ants Path length (Including the distance to the starting point); 9: endfor 10: Select the best ant in the current iteration and update the global ant. and ;11: For each edge ;12: From formula Calculate pheromone increment and update pheromone; 13: endfor; 14: return .

[0031] Furthermore, the obstacle avoidance algorithm based on the improved Rapidly-exploring Random Tree (RRT) is specifically as follows: After completing the initial path planning, each UAV path generated by the ant colony algorithm is formed by a straight line connecting two task points. However, in the actual combat environment, there may be obstacles or no-fly zones on the straight path. Therefore, it is necessary to optimize the straight path for local obstacle avoidance. The Rapidly-exploring Random Tree (RRT) algorithm is widely used in path planning due to its exploration capability in high-dimensional space. This algorithm can easily handle scenarios containing obstacles, but the paths it generates usually have redundant turning points and insufficient convergence speed. Therefore, an improved RRT algorithm that integrates target bias strategy and path pruning optimization is proposed. It achieves local obstacle avoidance by generating intermediate path points and shortens the path length as much as possible while ensuring safety. Problem description and algorithm construction: After task allocation and initial path planning, the final obstacle avoidance route planning problem remains for multi-UAV task allocation and path planning, that is, solving the obstacle avoidance path generation between two points. Assume that the UAV needs to start from the starting point To the finish line Planning obstacle avoidance paths, the map environment is composed of two-dimensional grids It means that, among them This refers to the vertical size of the raster map. To set the horizontal size of the raster map, use Define step size for identifying obstacle areas The target attraction threshold is the maximum distance for a single expansion. To determine the critical distance to the target; the algorithm's objective is to generate a path. , so that: (1) Located in free space; (2) (3) The total path length is minimized, i.e., the straight line segment does not intersect with obstacles; The algorithm and improvement strategies address the shortcomings of traditional fast exploration random tree algorithms by adding a target-biased expansion strategy. This introduces target point attraction during random sampling, accelerating path convergence. Efficient collision detection on discrete grid maps is achieved based on the Bressenham line algorithm. A path pruning optimization stage is added at the end of the algorithm, using line detection to remove redundant path points, improving path smoothness and economy. Hybrid directional expansion: Compared to the completely random expansion strategy used in traditional fast exploration random tree algorithms, in UAV mission allocation scenarios where target coordinates are determined, the target direction is prioritized as the exploration expansion direction, effectively improving the exploration convergence speed. First, a probability of expanding new nodes towards the target direction is set. ,make , For the target coordinates, , Calculate the expansion angle based on the coordinates of the node in the tree closest to the target. Then, candidate nodes are generated: At the same time, the original random expansion method is retained, based on probability. Random sampling in free space yields points. ,along arrive Nodes are expanded in the direction of collision detection: The Bressenham line algorithm is used to generate a discrete grid point set for the straight-line path between two points. To ensure that the straight-line grid between two points does not overlap with the obstacle grid and does not exceed the grid map boundary, efficient collision detection is achieved through the following formula: When any grid point crosses the boundary or becomes an obstacle, it returns to a collision state; path backtracking and pruning optimization: when a tree node enters the target area, it backtracks along the tree structure towards the parent node to generate the original path. A greedy strategy is used to prune the path; first, the optimized path is initialized. Current testing point ; then from Start, then traverse backwards. Choose the safest point that can be connected by a straight line. join in Last updated Repeat until the end is reached. .

[0032] After adopting the above technical solution, the beneficial effects of the present invention are as follows: The present invention aims to improve the efficiency of multi-UAV collaborative search and rescue missions, and provides a two-stage task allocation method and a two-stage path planning method. The two-stage task allocation method proposes a joint optimization framework combining mixed integer linear programming and improved ant colony algorithm, which can optimize task load balance and total flight distance, and achieve better task allocation effect. The two-stage path planning method proposes a two-stage path planning scheme that combines global task order optimization and local obstacle avoidance planning. By combining the global and local two-stage optimization scheme, a better path planning effect is achieved. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 The algorithm effect in this invention Figure 1 .

[0035] Figure 2 The algorithm effect in this invention Figure 2 .

[0036] Figure 3 This is the system deduction diagram in this invention (the right side of the diagram is the front-end task definition interface, and the left side is the back-end algorithm deduction interface). Detailed Implementation

[0037] See Figures 1-3 As shown, the technical solution adopted in this specific implementation is as follows: it includes a two-stage task allocation method and a two-stage path planning method. 1. Two-Stage Task Allocation Method (TADP): This method proposes a joint optimization framework combining Mixed Integer Linear Programming (MILP) and an improved ant colony algorithm. First, MILP is used to model the characteristics of multi-UAV combat missions, solving the model to achieve initial task allocation and addressing the task allocation problem for multiple UAVs starting from different points. Furthermore, considering that multiple UAVs often start from the same point in reality, a Single Warehouse Multiple Traveling Salesman Problem (SD-MTSP) model is constructed for UAV swarms starting from the same point. An elite strategy and a 2-opt local optimization design are then integrated to improve the ant colony algorithm for secondary task allocation, optimizing task load balancing and total flight distance for better task allocation performance. Its two-stage task assignment method (TADP) specifically addresses the task assignment problem in complex scenarios where multiple UAVs start from different points by combining mixed-integer linear programming (MILP) with an improved ant colony algorithm. This method consists of two core stages: The first stage involves global task allocation using mixed-integer linear programming: This stage rapidly allocates global tasks based on a mixed-integer linear programming model, optimizing total flight distance and mission integrity. Specifically, the global task allocation using mixed-integer linear programming is as follows: S1, Problem Modeling: Task allocation in the context of multiple drones performing search and rescue missions can typically be described as: drone swarm From each starting point Multiple destinations need to be reached. Execute the task At the same time, the following constraints must be met: (1) Task integrity: All tasks to be executed must be assigned and assigned only once; (2) Load constraint: The number of tasks assigned to a single UAV shall not exceed its maximum load capacity; (3) Distance Optimization: Minimize the total flight distance, considering minimizing the Euclidean distance from the UAV to the mission point; The drone formation includes Homogeneous drones Set of tasks to be executed ,in Indicates drone initial coordinates, Indicates task Location coordinates, This represents the maximum workload for a single machine. decision variables Defined as: With minimizing the total flight distance as the objective, construct the objective function: Among them, distance coefficient The calculation is as follows: Based on the practical requirements of multi-UAV task allocation, the following constraints are established: (1) Task allocation integrity constraint: This constraint ensures that each task It was assigned by and only by one drone.

[0038] (2) Unmanned Aerial Vehicle (UAV) payload capacity constraints: This constraint ensures that the number of tasks assigned to each drone does not exceed its maximum payload.

[0039] (3) Variable type constraints: From a feasibility perspective, to ensure the existence of a feasible solution for the system, the total payload capacity of the UAV must cover all mission loads, i.e. From the perspective of optimality guarantee, when the solver returns the globally optimal solution, the resulting allocation scheme reaches the theoretical optimum under given constraints. In practical engineering, an approximate optimal solution can be obtained by setting a time limit. S2, Model Solving Algorithm: (1) Input parameter preprocessing Initial coordinate matrix of the UAV: Task point coordinate matrix: Distance matrix calculation: .

[0040] (2) Construction of constraint matrix Let the equality constraint matrix Its elements satisfy: Among them, the index of the decision variable column Each row corresponds to one task, ensuring that the task is assigned only once; let the inequality constraint matrix... Each row corresponds to one drone, which is used to constrain the total number of missions.

[0041] (3) Optimization solution Transform the problem into the standard mixed-integer linear programming (MILP) form: in , , The flattened distance vector .

[0042] The Highs solver is used to solve the branch and bound algorithm, and a convergence tolerance is set. To ensure calculation accuracy.

[0043] The second stage involves improving the local optimization of the ant colony algorithm: A single-warehouse multiple traveling salesman problem (SD-MTSP) model is constructed for drone swarms starting from the same point. By introducing an elitist strategy, 2-opt local optimization, and a dynamic load balancing mechanism, the task load balance and total flight distance are further optimized. Specifically, the improved local optimization of the ant colony algorithm includes: S1, Problem Modeling: After the initial assignment is completed in mixed-integer programming, for the same starting point... The problem of secondary task allocation for drones can be modeled as a single-warehouse multiple traveling salesman problem (SD-MTSP). Given a shared initial location... The drone formation, including Homogeneous drones Set of tasks to be executed coordinate set Its constraints are as follows: (1) Each task point is visited only once; (2) All drones return to their initial positions. (3) Minimize total flight distance This problem can be formalized as: in Indicates the first The closed-loop path of the drone. express , The Euclidean distance between them; S2, Algorithm Design: Based on the traditional ant colony algorithm, and taking advantage of the characteristics of SD-MTSP, an improved ant colony algorithm is designed, which includes the following core components: (1) Pheromone dynamic update mechanism An elite strategy is adopted, where only the globally optimal path participates in pheromone enhancement, and edges are defined. pheromone concentration Update rules: in, The pheromone evaporation coefficient; the pheromone concentration increment. , Let pheromone intensity be constant. The shortest path in this round The total path length of a single ant; (2) Path selection probability model Ants at the node Select the next node The probability of: in, As a heuristic factor; Controlling the influence weight of pheromones Control the weight of heuristic information; (3) Load balancing strategy To avoid uneven task distribution, dynamic load constraints are introduced: in For the set threshold, The total number of tasks. For the first The number of tasks assigned to each drone is achieved by dynamically adjusting the path length during the ant colony search process; (4) Local optimization using the 2-opt algorithm The 2-opt algorithm is a local search algorithm for optimizing paths. It improves the quality of a path by swapping two nodes in the path. Ant colony optimization acts as a global planner, constructing diverse solutions; while 2-opt acts as a local planner, enhancing the solution in the local search phase by refining the path and eliminating inefficient sub-paths. Through the cooperation of global and local algorithms, the task allocation results are optimized.

[0044] 2. Two-Stage Path Planning Method (PPDP): This method proposes a two-stage path planning scheme that combines global task order optimization with local obstacle avoidance planning. The first stage uses the ant colony algorithm to solve the Traveling Salesman Problem (TSP) and optimize the global access order of the UAV to task points. The second stage combines an improved Rapid Random Tree (RRT) algorithm with target bias strategy and path pruning optimization to generate obstacle avoidance paths that balance safety and economy. By combining global and local two-stage optimization, a better path planning effect is achieved. The specific details of the two-stage path planning method (PPDP) are as follows: In multi-UAV collaborative search and rescue missions, path planning is particularly critical. Relying on a single planning algorithm makes it difficult to simultaneously achieve global path order optimization and local obstacle avoidance. To address this, the PPDP (Path Planning with Dual-Phase Optimization) scheme is proposed. This method achieves an organic combination of global task order optimization and local collision and obstacle avoidance through two-stage collaborative optimization. The first stage involves using ant colony optimization (ACO) to solve the Task Point Solution (TSP) problem, optimizing the UAV's access order to various task points. Specifically, the ACO algorithm addresses the TSP problem as follows: After multi-drone task allocation is completed, each drone has a clearly defined set of task points, thus significantly reducing the complexity of path planning. At this point, the multi-drone path planning problem can be equivalent to multiple independent Traveling Salesman Problems (TSPs), with each TSP instance corresponding to all the task points a drone needs to visit. For this type of path optimization problem, the Ant Colony Optimization (ACO) algorithm is widely used in the field of path planning due to its strong global search capability and adaptability. In ant colony optimization, each ant starts from a node and selects the next node to visit based on probability rules, until it has visited all nodes and returns to the starting point. The quality of this path guides the updating of pheromones: the shorter the path, the higher the pheromone concentration along it, thus attracting subsequent ants to choose that path first.

[0045] S1, Problem Description and Model Building: Let the set of mission points for a certain drone be . ,in This represents the starting point of the drone, and the rest are the assigned task points. Define the distance matrix. ,in Represents a node With nodes Euclidean distance between them: in, For a minimum value (e.g.) To avoid the abnormal situation of division by zero; Define the pheromone matrix heuristic function Pheromones influence weighting factors Heuristic information weighting factor pheromone volatility coefficient pheromone intensity constant Pheromones concentration increment and dynamic pheromone update mechanism.

[0046] S2, Solution Algorithm: In each iteration, all ants complete the entire path once, calculate and record the path length, and update the optimal path and global pheromone. To improve convergence speed, an elite strategy is introduced in the algorithm: the optimal path (such as the globally optimal or iteratively optimal ant) ​​is given a higher pheromone increment, as detailed in Algorithm 1. Algorithm 1 is as follows: The second stage involves an obstacle avoidance algorithm based on an improved Fast Random Tree (RRT) algorithm. This algorithm generates a safe path that considers obstacles, ensuring the UAV completes its mission while avoiding them. This two-stage approach minimizes the UAV's path length and significantly improves the smoothness and safety of path planning. Specifically, the obstacle avoidance algorithm based on the improved RRT algorithm is as follows: After completing the initial path planning, each UAV path generated by the ant colony algorithm is formed by connecting two task points with a straight line. However, in the actual combat environment, there may be obstacles or no-fly zones on the straight path. Therefore, it is necessary to optimize the straight path for local obstacle avoidance. The Rapidly-exploring Random Tree (RRT) algorithm is widely used in path planning due to its exploration capability in high-dimensional space. This algorithm can easily handle scenarios containing obstacles, but the paths it generates usually have redundant turning points and insufficient convergence speed. To address this, an improved RRT algorithm that integrates target bias strategy and path pruning optimization is proposed. This algorithm achieves local obstacle avoidance by generating intermediate path points and shortens the path length as much as possible while ensuring safety.

[0047] S1, Problem Description and Algorithm Construction: After task allocation and preliminary path planning, the final problem in multi-UAV task allocation and path planning is obstacle avoidance route planning, which involves generating an obstacle avoidance path between two points. Assume the UAV needs to start from the starting point... To the finish line Planning obstacle avoidance paths, the map environment is composed of two-dimensional grids It means that, among them This refers to the vertical size of the raster map. To set the horizontal size of the raster map, use Define step size for identifying obstacle areas The target attraction threshold is the maximum distance for a single expansion. To determine the critical distance to reach the target; The algorithm's goal is to generate paths. , so that: (1) Located in free space; (2) (3) The total path length is minimized, i.e., the straight line segment does not intersect with obstacles; .

[0048] S2, Solution Algorithm and Improvement Strategies: To address the shortcomings of traditional fast exploration random tree algorithms, a target bias expansion strategy is added to introduce target point attraction during random sampling, accelerating path convergence. Efficient collision detection for discrete grid maps is achieved based on the Bressenham line algorithm. Finally, a path pruning optimization stage is added to the algorithm to remove redundant path points using line detection, improving path smoothness and economy. (1) Mixed direction expansion Compared to the traditional fast exploration random tree algorithm which uses a completely random expansion direction strategy, in the UAV mission allocation scenario, where the target coordinates are determined, we consider prioritizing the target's direction as the exploration expansion direction, which can effectively improve the exploration convergence speed.

[0049] First, set a probability for expanding new nodes in the target direction. ,make , For the target coordinates, , Calculate the expansion angle based on the coordinates of the node in the tree closest to the target. Then, candidate nodes are generated: At the same time, the original random expansion method is retained, based on probability. Random sampling in free space yields points. ,along arrive Expand nodes in the direction of [the target].

[0050] (2) Collision Detection The Bressenham line algorithm is used to generate a discrete grid point set for a straight path between two points. To ensure that the straight-line grid between two points does not overlap with the obstacle grid and does not exceed the grid map boundary, efficient collision detection is achieved through the following formula: When any grid point crosses the boundary or becomes an obstacle, it returns to the collision state. (3) Path backtracking and pruning optimization Once a tree node enters the target region, it traces back along the tree structure towards its parent node to generate the original path. A greedy strategy is used to prune the path; First, initialize the optimization path. Current testing point ; then from Start, then traverse backwards. Choose the safest point that can be connected by a straight line. join in Last updated Repeat until the end is reached. .

[0051] The above description is only used to illustrate the technical solution of the present invention and is not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention, as long as they do not depart from the spirit and scope of the technical solution of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A two-layer, two-stage heterogeneous UAV task allocation and trajectory planning method, characterized in that: It includes a two-stage task allocation method and a two-stage path planning method: 1) Two-stage task allocation method: a joint optimization framework combining mixed integer linear programming and improved ant colony algorithm is proposed. First, a mixed integer linear programming model is established according to the characteristics of multi-UAV combat tasks, and the initial task allocation is solved. Considering the actual situation that multiple UAVs usually exist at the same starting point, a multi-traveling salesman problem model is constructed for the UAV group at the same starting point, and an improved ant colony algorithm is designed by combining elite strategy and 2-opt local optimization to perform secondary task allocation, optimize task load balance and total flight distance, and achieve better task allocation effect. 2) Two-stage path planning method: a two-stage path planning scheme combining global task order optimization and local obstacle avoidance planning is proposed. In the first stage, the traveling salesman problem is solved by using the ant colony algorithm to optimize the global access order of UAVs to task points. In the second stage, the improved rapid exploration random tree algorithm is combined, the target bias strategy and path pruning optimization are introduced, and the obstacle avoidance path considering safety and economy is generated. Through the combination of global and local two-stage optimization scheme, better path planning effect is achieved. The two-stage task allocation method specifically includes:

2. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method according to claim 1, wherein: First stage, mixed integer linear programming global task allocation: based on the mixed integer linear programming model, the global task allocation is quickly completed, and the total flight distance and task integrity are optimized. Second stage, improved ant colony algorithm local optimization: a single warehouse multi-traveling salesman problem model is constructed for the UAV group at the same starting point, and the task load balance and total flight distance are further optimized by introducing elite strategy, 2-opt local optimization and dynamic load balancing mechanism. The mixed integer linear programming global task allocation specifically includes:

3. The bi-layer bi-stage heterogeneous UAV mission assignment and trajectory planning method according to claim 2, wherein: S1, problem modeling: (1) Task integrity: all tasks to be executed must be assigned and assigned only once; Task allocation in the context of multi-UAV search and rescue missions can generally be described as follows: a set of UAVs from each starting point need to go to multiple target points , perform a task while satisfying the following constraints: (2) Load constraint: the number of tasks assigned to a single UAV does not exceed its maximum load capacity; (3) Distance optimization: total flight distance minimization, considering the Euclidean distance minimization from the starting point to the task point; (1) Task allocation integrity constraint: The unmanned aerial vehicle formation comprises The same structure unmanned aerial vehicle A set of tasks to be executed Wherein The initial coordinates of the unmanned aerial vehicle , The position coordinates of the task , The maximum single task load The decision variables are defined as: x = (x1, x2, x3, x4, x5, A target function is constructed with the goal of minimizing the total flight distance: wherein the distance coefficient is calculated as: Based on the actual requirements of multi-UAV task allocation, the constraints are constructed as follows: (2) UAV load capacity constraint: This constraint ensures that each task is assigned to and only to one drone; (3) Variable type constraint: This constraint ensures that the number of tasks assigned to each drone does not exceed its maximum load; S2, model solving algorithm: From the feasibility condition, to ensure the system has a feasible solution, the total payload capacity of UAVs should cover all task quantities, i.e. From the optimality guarantee, when the solver returns the global optimal solution, the obtained allocation scheme reaches the theoretical optimum under the given constraints. In practical engineering, an approximate optimal solution can be obtained by setting an upper limit on the time. (1) Input parameter preprocessing UAV initial coordinate matrix: (3) Optimization solution waypoint coordinate matrix: Distance matrix calculation: (2) Constraint matrix construction Let the equality constraint matrix whose elements satisfy: where the decision variable column index Each row corresponds to a task, ensuring that the task is assigned only once. Inequality constraint matrix Each row corresponds to a UAV, which is used to constrain its total number of tasks; Convert the problem to a standard mixed integer linear programming form: The improved ant colony algorithm local optimization specifically includes: wherein, , , is the flattened distance vector, ; The Highs solver is used for the branch and bound algorithm, and a convergence tolerance is set. To ensure calculation accuracy.

4. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method of claim 2, wherein: (1) Each task point is visited only once; S1, problem modeling: after the initial allocation is completed by the mixed integer programming, the secondary task allocation problem of the UAVs is modeled as a single warehouse multi-traveler problem, given that the UAV formation shares the same initial location with constraints:​​​​​ (2) All UAVs return to the initial position (3) Total flight distance minimization This problem can be formalized as: S2, algorithm design: wherein, represents the closed loop path of the drone, represents , Euclidean distance between Based on the traditional ant colony algorithm, an improved ant colony algorithm is designed based on the characteristics of the single warehouse multi-traveling salesman problem, which includes the following core components: (1) Information dynamic updating mechanism (2) Path selection probability model The elite strategy is adopted, only the global optimal path participates in pheromone enhancement, and the pheromone concentration of the edge is defined The update rule: wherein, is the pheromone volatility coefficient; the pheromone concentration increment , is the pheromone intensity constant, is the shortest path of the current round, total path length of the only ant; (3) Load balancing strategy Ants at nodes Selecting the probability of the next node : wherein, is a heuristic factor; control pheromone influence weight, control heuristic information weight; To avoid uneven task allocation, dynamic load constraints are introduced: (4) 2-opt algorithm is used for local optimization wherein, is a set threshold value, is the total number of tasks, is the first The number of tasks allocated to the drone is achieved by dynamically adjusting the path length during the ant colony search process. ​ 2-opt algorithm is a local search algorithm for optimizing path, which improves the quality of path by exchanging two nodes in the path, and the ant colony algorithm is used as a global planner to build diversified solutions; 2-opt is used as a local planner to enhance the solution by refining the path and eliminating inefficient sub-paths in the local search stage; through the cooperation of global and local, the task allocation result is optimized.

5. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method of claim 1, wherein: The two-stage path planning method is specifically: In the multi-unmanned aerial vehicle cooperative search and rescue task, the path planning problem is particularly important, therefore, a two-stage path planning method is proposed, which realizes the organic combination of global task sequence optimization and local collision avoidance through two-stage cooperative optimization, In the first stage, the ant colony algorithm solves the traveling salesman problem: the ant colony algorithm is used to solve the traveling salesman problem to optimize the visiting sequence of each task point by the unmanned aerial vehicle; In the second stage, the improved rapid exploration random tree algorithm is used to generate a safe path considering obstacles to ensure that the unmanned aerial vehicle completes the task while avoiding obstacles, which not only ensures that the path length of the unmanned aerial vehicle is as short as possible, but also greatly improves the smoothness and safety of the path planning.

6. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method of claim 5, wherein: The ant colony algorithm for solving the traveling salesman problem is specifically: After completing the multi-unmanned aerial vehicle task allocation, each unmanned aerial vehicle has determined its corresponding task point set, and the complexity of path planning is greatly reduced, at this time, the multi-unmanned aerial vehicle path planning problem can be equivalent to multiple independent traveling salesman problems, each traveling salesman problem instance corresponds to all the task points that need to be visited by an unmanned aerial vehicle, for this kind of path optimization problem, the ant colony algorithm is widely used in path planning field due to its strong global search ability and adaptability; In the ant colony algorithm, each ant starts from a certain node, selects the next visited node based on the probability rule, and returns to the starting point after visiting all nodes, the quality of the path is used to guide the update of pheromone: the shorter the path, the higher the concentration of pheromone distributed on it, thereby attracting subsequent ants to preferentially select; S1, problem description and model construction: Let the set of mission points of a UAV be wherein denotes the departure point of the UAV, and the rest are assigned mission points, and define the distance matrix wherein denotes the Euclidean distance between node and node . wherein, is a minimum value to avoid an exception of division by zero; Definition of pheromone matrix , Heuristic function , Pheromone influence weight factor , Heuristic information weight factor , Pheromone evaporation coefficient , Pheromone intensity constant , Pheromone concentration increment, Pheromone dynamic updating mechanism S2, solution algorithm: In each iteration, all ants complete a complete path, calculate and record the path length, and update the optimal path and global pheromone, in order to improve the convergence speed, the elite strategy is introduced in the algorithm: the optimal path is given a higher pheromone increment, the specific algorithm is shown in algorithm 1.

7. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method of claim 6, wherein: The algorithm 1 is specifically: Ant colony algorithm processing traveling salesman problem: input: , , , , , , set of coordinates of unmanned aerial vehicle task points ; Output: optimal path node sequence ; initialization: by formula calculate distance matrix pheromone matrix, pheromone matrix , heuristic function global optimal path , length ; 1: for to do; 2: for each Ant to do; 3: Randomly select a starting node, construct an initial tabu list 4: for to do; 5: by formula Compute transition probabilities from current node to all unvisited nodes set ; 6: randomly select next node according to probability distribution, and add ; 7: end for; 8: compute ants path length 9: endfor 10: select the best ant in the current iteration, update the global best and ; 11: for each edge ; 12: compute pheromone increment and update pheromone; 13: end for; compute pheromone increment and update pheromone; 13: end for; 14: return .

8. The dual-layer dual-stage heterogeneous UAV mission assignment and trajectory planning method of claim 5, wherein: The obstacle avoidance algorithm based on the improved rapid exploration random tree is specifically: After the initial path planning, each path generated by the ant colony algorithm is connected by a straight line between two task points. However, there may be obstacles or no-fly zones on the straight line in the actual combat environment, so it is necessary to optimize the local obstacle avoidance of the straight line path. The rapidly-exploring random tree algorithm is widely used in path planning due to its exploration ability in high-dimensional space. This algorithm can easily handle scenarios with obstacles, but the generated path often has redundant turning points and insufficient convergence speed. Therefore, an improved RRT algorithm is proposed, which combines target bias strategy and path pruning optimization. The algorithm generates intermediate path points to achieve local obstacle avoidance and shortens the path length as much as possible under the premise of safety. S1, problem description and algorithm construction: After task assignment and preliminary path planning, the multi-UAV task assignment and path planning is left with the last obstacle avoidance route planning problem, i.e. solving the obstacle avoidance path generation between two points, assuming that the UAV needs to plan an obstacle avoidance path from the starting point to the end point , the map environment is represented by a two-dimensional grid , where is the longitudinal size of the grid map, is the transverse size of the grid map, using to identify the obstacle area, the step size is defined as the maximum distance of a single expansion, and the target attraction threshold is the critical distance to determine the arrival of the target; The algorithm aims to generate paths such that: (1) Located in free space; (2) The straight line segment does not intersect the obstacle; (3) the total length of the path is minimized, i.e. ; S2, solution algorithm and improvement strategy: To overcome the shortcomings of the traditional rapidly-exploring random tree algorithm, a target bias expansion strategy is added to introduce target point attraction during random sampling, accelerating path convergence. Based on the Breitenham straight line algorithm, efficient collision detection of discrete grid maps is achieved. Finally, path pruning optimization is added to remove redundant path points using straight line detection, improving path smoothness and economy. (1) Mixed direction expansion Compared to the completely random expansion direction strategy of the traditional rapidly-exploring random tree algorithm, in the unmanned aerial vehicle task allocation scenario, the target direction is considered as the preferred exploration expansion direction when the target coordinates are determined. This can effectively improve the exploration convergence speed. First, set a probability of expanding a new node towards the target direction , let , be the target coordinate, , the coordinate of the node closest to the target in the tree, calculate the expansion angle , then generate the candidate node: At the same time, the original random expansion mode is retained, and the probability In free space, random sampling is performed to obtain points , and the nodes are expanded along the direction of to . (2) Collision detection generating a discrete grid of points along a straight line path between two points using the brezenham straight line algorithm , ensuring that no straight line grid between the two points overlaps with the obstacle grid and does not exceed the grid map boundaries, i.e. enabling efficient collision detection by: where any grid point that goes out of bounds or is an obstacle returns a collision state; (3) Path backtracking and pruning optimization When the tree node enters the target area, the original path is generated by backtracking along the tree structure towards the parent node , and the path is pruned by using the greedy strategy First, initialize the optimization path , the current detection point ; then from , traverse backwards, select the farthest safe point that can be connected by a straight line to join ; finally, update , repeat until the end point is reached .

Images