Unmanned aerial vehicle cluster path planning method based on self-adaption

Abstract

The invention relates to the field of unmanned aerial vehicle cluster path planning, and provides a remote upgrading method of an intelligent electric meter, which comprises the steps of 1, flight environment modeling, 2, algorithm parameter setting and particle initialization, 3, fitness evaluation, 4, external archive updating, 5, particle speed and position updating, 6, adaptive mutation operation, and 7, iteration termination judgment. And 8, optimal path generation and instruction conversion. When the grid method is adopted, the grid side length is dynamically adjusted, higher resolution is adopted in a high-risk area of a no-fly zone, the modeling precision of a key area is improved, the grid method and a Voronoi diagram method are combined, modeling modes are switched according to task requirements, and the environmental adaptability is enhanced; point cloud data are obtained through a laser radar and a visual sensor, the grid state is updated in real time through Kalman filtering or particle filtering, and timely recognition of a dynamic obstacle is ensured.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm path planning, and more specifically to an adaptive UAV swarm path planning method. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarms have broad application prospects in fields such as military reconnaissance, disaster relief, and logistics delivery. Path planning is one of the key technologies for collaborative control of UAV swarms. Its purpose is to plan an optimal or suboptimal path from the starting point to the target point for each UAV, while satisfying constraints such as obstacle avoidance, collaboration, and energy consumption.

[0003] In traditional UAV swarm path planning, environmental modeling often uses fixed-resolution grid methods or single Voronoi diagrams, which cannot dynamically adjust the resolution of local areas according to task requirements, resulting in low computational efficiency or insufficient modeling accuracy in areas with dense obstacles.

[0004] Existing algorithms often focus on a single objective (such as the shortest path or the lowest energy consumption) and lack comprehensive optimization of multiple objectives such as safety and time, resulting in one-sided path planning results.

[0005] Traditional particle swarm optimization (PSO) is prone to getting trapped in local optima and lacks an adaptive mechanism, resulting in inefficient path planning.

[0006] The generated waypoints may contain sharp turns or not meet the speed / acceleration limits of the drone, causing the execution to fail.

[0007] Traditional methods struggle to handle obstacles that change in real time (such as sudden drone attacks or moving vehicles). To address these issues, we propose an adaptive drone swarm path planning method. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an adaptive UAV swarm path planning method. This invention systematically solves the problems of rigid modeling, one-sided optimization, poor convergence, inconsistency with physical constraints, and insufficient dynamic adaptability in traditional UAV swarm path planning by using dynamic modeling, multi-objective optimization, adaptive mutation, path smoothing constraints, and local update mechanisms, thus significantly improving the efficiency and reliability of path planning.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] An adaptive UAV swarm path planning method includes the following steps:

[0011] Step 1: Flight Environment Modeling. After receiving the path planning task trigger signal, the first step is to model the UAV's flight environment. After modeling, the environmental information is converted into a computer-readable digital map.

[0012] Step 2: Algorithm parameter setting and particle initialization: After completing the environment modeling, set the parameters of the particle swarm optimization algorithm. Then, randomly initialize the position and velocity of the particles. At the same time, initialize the external archive to an empty set to store non-dominated solutions.

[0013] Step 3 Fitness Evaluation: Based on the constructed digital map and multi-objective optimization requirements, calculate the fitness value of the path scheme represented by each particle in the current population;

[0014] Step 4: Update the external archive: Add non-dominated solutions from the current population that are not dominated by any archived solution to the archive, and remove old solutions dominated by newly added solutions; if the archive size exceeds the preset capacity, use the crowding distance sorting strategy to eliminate inferior solutions in the "crowded region".

[0015] Step 5: Particle velocity and position update: Based on the individual historical optimal solution of each particle, update the velocity of each particle according to subgroup optimization;

[0016] Step 6: Adaptive Mutation Operation: Based on the current iteration number and particle fitness value, dynamically adjust the mutation probability and mutation magnitude, and perform mutation operations on some particles.

[0017] Step 7: Iteration Termination Judgment: If the termination condition is met, the algorithm ends and the optimal solution in the external archive is output as the final result; otherwise, return to step 3 to continue execution.

[0018] Step 8: Optimal Path Generation and Command Conversion: Select the optimal path scheme from the final results and convert it into flight commands that the UAV can execute.

[0019] Preferably, the modeling method adopts the grid method, which divides the flight area into uniform regular grids, and the resolution is dynamically adjusted according to the mission requirements; the formula is: grid side length s=L / N, where L is the side length of the area and N is the number of grids;

[0020] Obstacle point cloud data is acquired using LiDAR and visual sensors. If an obstacle point cloud exists within the grid, it is marked as an "obstacle grid"; otherwise, it is marked as a "passable grid".

[0021] Preferably, the modeling method employs the Voronoi diagram method, which uses obstacle vertices or center points as generators to ensure coverage of all obstacles; equidistant lines of the Voronoi boundary are generated point-by-point using scan lines, where Pi( , ) and Pj( , The equation of the perpendicular bisector of () is: The drone's path is within the Voronoi region, maintaining a set safe distance from obstacles; when new obstacles are added, the Voronoi map of the affected area is recalculated.

[0022] Digital maps collect three-dimensional data of the flight environment through sensors, including the location, shape, and size of obstacles; the collected data is preprocessed, including noise reduction, smoothing, and segmentation operations; and the processed data is converted into a digital map according to the selected modeling method; the generated digital map is stored in a computer database, which supports subsequent algorithm calls.

[0023] Preferably, the parameters of the algorithm include, but are not limited to, population size, maximum number of iterations, inertia weight, and learning factor; each particle corresponds to a path planning scheme for a drone swarm, transforming the path problem into a numerical encoding of decision variables;

[0024] Particle position initialization: First, assume there are M drones in the cluster, and each drone's path consists of K nodes. Then, the dimension of a single particle is D = M × K × 2. The particle position is randomly generated within the path coordinate boundaries, expressed as: ,in, Indicates the first The particle in the first The initial position in each dimension, where Represents the particle number. Represents the dimension number; Indicates the first Minimum boundary value in each dimension Indicates the first Maximum boundary value in each dimension This represents a random number uniformly distributed in the interval [0,1], used to generate random positions for particles;

[0025] If the initial position corresponds to a drone path that traverses obstacles, adjustments are necessary. These adjustments can be made by randomly perturbing and fine-tuning the point coordinates or by regenerating the values ​​in that dimension, ensuring that the adjusted initial solution satisfies the basic constraints of the drone's flight.

[0026] The velocity determines the particle's search direction and step size. The initial velocity needs to be small to avoid jumping out of the feasible region. The expression for initializing the particle velocity is: ,in, Indicates the first The particle in the first Initial velocity in each dimension Indicates the first The particle in the first The minimum velocity in the dimension, Indicates the first The particle in the first The maximum value of the velocity in the dimension, This represents a random number uniformly distributed in the interval [0,1], used to randomly generate an initial velocity between the minimum and maximum values, ensuring the randomness and diversity of the velocity.

[0027] Initialize external archive A= After generating the initial particle swarm, the multi-objective fitness value of each particle is calculated, and the initial non-dominated solutions are selected and stored through non-dominated sorting. The archive maintenance mechanism is dynamically updated in subsequent iterations to ensure that it stores the current optimal and diverse set of solutions.

[0028] Preferably, the fitness value expression is: ,in, The fitness value is used to measure the quality of the path scheme represented by each particle in the current population. It represents the distance from the starting point to the ending point; This represents the energy consumed by the proposed path. Indicators representing the safety of the route plan. Indicates the time required to complete the route plan. , , , The represents the weighting coefficient, used to balance the relationship between the four objectives of path length, energy consumption, safety, and time.

[0029] Preferably, the newly added non-dominated solutions are screened: For each particle in the current population, its dominance relationship with all solutions in the external archive is calculated. If the particle is not dominated by any solution in the archive, it is added to the external archive. The specific process is as follows:

[0030] For each solution q in the current particle and the external archive A, determine whether q dominates the current particle; if it is not dominated by any solution in the archive, add it to the external archive; otherwise, do not add it. After adding a new solution to the external archive, old solutions dominated by the newly added solution need to be removed. For each solution in the archive, determine whether there is a newly added solution that dominates it; if so, remove it from the archive. When the capacity of the external archive reaches its maximum limit, a crowding sorting strategy is used to eliminate inferior solutions in "crowded regions" to ensure the diversity of solution distribution within the archive.

[0031] For each solution, calculate its crowding degree in the target space; the formula for calculating the crowding degree is: ,

[0032] in, ‎ represents the solution The degree of congestion is used to measure the degree of congestion of the solution. The relative density in the target space; Representing the solution Its adjacent solutions In the The difference in values ​​on each objective function; This indicates the number of objective functions, i.e., how many optimization objectives need to be considered; Representing the solution In the The values ​​of each objective function; Representing the solution Adjacent solutions The value of the i-th objective function;

[0033] Sort the solutions in the archive in descending order according to their crowding level, retain the N solutions with the highest crowding level, and discard the rest.

[0034] Preferably, the speed update expression is: ,in, Represents particles In dimensions The speed on, Represents particles In dimensions Current position on Represents particles In dimensions The best historical position on Represents all particles in dimension The global optimal position on , A random number between [0,1] is used to increase the randomness and diversity of the algorithm; Inertial weights are used to balance the influence of a particle's historical velocity on its current velocity. and These are cognitive learning factors and social learning factors, which respectively regulate the degree to which particles learn towards individual and global optima.

[0035] Preferably, the mutation probability decreases as the number of iterations increases, and the specific calculation formula is as follows: ,in, The mutation probability is the number of iterations. The initial mutation probability, This represents the current iteration number. The maximum number of iterations,

[0036] The mutation amplitude decreases as the particle fitness value increases, and the specific calculation formula is as follows: ,in, This represents the variation range of the current particle. The initial variation amplitude, This represents the current particle's fitness value. This represents the maximum fitness value in the current population.

[0037] For each particle, generate a random number r. If r < 0, then... Then, a mutation operation is performed on the particle, the specific formula of which is: ,in, For the first The first particle Dimensional position, These are random numbers that follow a standard normal distribution.

[0038] Preferably, the iteration termination judgment is as follows: if the maximum number of iterations T is reached or the fitness value has not been significantly improved for N consecutive iterations, the loop is exited and step 8 is entered; otherwise, step 3 is returned; the non-dominated solution with the highest comprehensive score is selected from the external archive as the final optimal path.

[0039] Preferably, the optimal path selection involves normalizing path length, energy consumption, and safety from the multi-objective solutions, and assigning weights based on task priority.

[0040] Calculate the overall score for each non-dominated solution using the weighted sum method: ,in, For the target weight, The normalized target value is used; the solution with the highest comprehensive score is selected as the final optimal path.

[0041] Flight command conversion path smoothing: Cubic spline interpolation is used to smooth the path points to ensure path continuity; the smoothed path is checked to see if it meets the speed limit, acceleration limit and endurance constraint of the UAV. If it does not meet the constraint, the path is readjusted; if it meets the constraint, the smoothed path is discretized into a sequence of waypoints at fixed intervals. Each waypoint contains coordinate, speed and attitude information for UAV flight control. The waypoint sequence is packaged into a command packet using a communication protocol and sent to the UAV flight control system via a wireless link.

[0042] This invention provides an adaptive UAV swarm path planning method. It offers the following advantages: When using the grid method, the grid edge length is dynamically adjusted, and higher resolution is used in high-risk areas such as no-fly zones, improving the modeling accuracy of key areas. Combining the grid method with the Voronoi diagram method allows switching between modeling modes according to task requirements, enhancing environmental adaptability. Point cloud data is acquired through LiDAR and visual sensors, and the grid state is updated in real time using Kalman filtering or particle filtering, ensuring timely identification of dynamic obstacles.

[0043] Define fitness values ​​to balance path length, energy consumption, safety, and time, and dynamically adjust priorities through weight coefficients; adopt non-dominated sorting and congestion distance strategies to retain diverse Pareto optimal solutions and avoid the limitations of a single solution; in the final path selection, redistribute weights according to task type (such as reconnaissance focusing on safety, transportation focusing on energy consumption) to achieve flexible decision-making.

[0044] The dynamic mutation probability decreases with increasing iterations, balancing global search and local optimization. Fitness drives the mutation amplitude, reducing the mutation amplitude of particles with high fitness to avoid excessive perturbation of high-quality solutions. Velocity and position update optimization employs the standard PSO velocity formula combined with inertia weights and learning factors to enhance search capabilities.

[0045] Cubic spline interpolation smoothing: Smooths the path points to ensure path continuity. Dynamic constraint checking: Checks whether the smoothed path meets speed, acceleration, and endurance time constraints. If not, it is resolved by readjusting the path or optimizing the waypoint sequence.

[0046] Local Voronoi map update: When an obstacle is added, only the Voronoi boundary of the affected area is recalculated, rather than the entire map, reducing computational overhead; Real-time sensor fusion: Digital maps are updated by preprocessing (denoising and segmentation) 3D data from LiDAR and cameras, combined with Kalman filtering, to achieve rapid response to dynamic environments. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the process of the present invention; Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] See attached document Figure 1 As shown, an adaptive UAV swarm path planning method includes the following steps:

[0050] Step 1: Flight Environment Modeling. After receiving the path planning task trigger signal, the first step is to model the UAV's flight environment. After modeling, the environmental information is converted into a computer-readable digital map.

[0051] The modeling method employs a grid-based approach, which divides the flight area into uniform, regular grids (such as squares or hexagons). The resolution is dynamically adjusted according to mission requirements. The formula is: grid side length s = L / N, where L is the side length of the area and N is the number of grids. Depending on obstacle density or mission priority, higher resolution can be used in local areas (such as around no-fly zones).

[0052] Obstacle point cloud data is acquired through LiDAR and visual sensors. If an obstacle point cloud exists within the grid (e.g., its height exceeds a threshold), it is marked as an "obstacle grid"; otherwise, it is marked as a "passable grid". Real-time sensor data is used to update the grid state through Kalman filtering or particle filtering.

[0053] The modeling method employs the Voronoi diagram method, which uses obstacle vertices or center points as generators to ensure coverage of all obstacles. It generates equidistant lines of the Voronoi boundary point-by-point using scan lines, where Pi(…)… , ) and Pj( , The equation of the perpendicular bisector of () is: The drone's path is within the Voronoi region, maintaining a set safe distance from obstacles; when a new obstacle is added, the Voronoi map of the affected area is recalculated (local update).

[0054] Digital maps collect 3D data of the flight environment through sensors (such as LiDAR and cameras), including the location, shape, and size of obstacles. The collected data undergoes preprocessing, including noise reduction, smoothing, and segmentation, to improve accuracy and consistency. Depending on the chosen modeling method (raster or Voronoi diagram), the processed data is converted into a digital map (e.g., in raster modeling, obstacle areas are marked as 0, and passable areas as 1; in Voronoi diagram modeling, each Voronoi region is mapped to a grid cell, and its status as an obstacle is marked). The generated digital map is stored in a computer database (the storage format is selected based on the modeling method: raster modeling uses a two-dimensional array (e.g., obstacle rasters are marked as 0, and passable rasters as 1); Voronoi diagram modeling uses a graph structure (nodes represent Voronoi vertices, and edges represent boundaries)). The digital map database supports subsequent algorithm calls (e.g., raster arrays are used for A* search, and Voronoi diagrams are used for the RRT* algorithm).

[0055] Step 2: Algorithm parameter setting and particle initialization: After completing the environment modeling, set the parameters of the particle swarm optimization (PSO) algorithm. Then, randomly initialize the position and velocity of the particles. At the same time, initialize the external archive to an empty set to store non-dominated solutions.

[0056] The parameters of the algorithm include, but are not limited to, population size, maximum number of iterations, inertia weight, and learning factor; each particle corresponds to a path planning scheme for a drone swarm, transforming the path problem into a numerical encoding of decision variables;

[0057] Particle position initialization (representing the path planning scheme): First, assume there are M drones in the cluster, and each drone's path consists of K nodes (including the start and end points). Then, the dimension of a single particle is D = M × K × 2 (i.e., the horizontal and vertical coordinates of each node). Randomly generate particle positions within the path coordinate boundaries, expressed as: ,in, Indicates the first The particle in the first The initial position in each dimension, where Represents the particle number. Represents the dimension number; Indicates the first Minimum boundary value in each dimension Indicates the first Maximum boundary value in each dimension This represents a random number uniformly distributed in the interval [0,1], used to generate random positions for particles;

[0058] If the initial position corresponds to the drone's path crossing obstacles, adjustments are required. These adjustments can be made by randomly perturbing the coordinates of the points or by regenerating the values ​​of that dimension to ensure that the adjusted initial solution meets the basic constraints of drone flight, such as avoiding collisions and complying with flight restrictions.

[0059] The velocity determines the particle's search direction and step size. The initial velocity needs to be small to avoid jumping out of the feasible region. The expression for initializing the particle velocity is: ,in, Indicates the first The particle in the first Initial velocity in each dimension Indicates the first The particle in the first The minimum velocity in the dimension, Indicates the first The particle in the first The maximum value of the velocity in the dimension, This represents a random number uniformly distributed in the interval [0,1], used to randomly generate an initial velocity between the minimum and maximum values, ensuring the randomness and diversity of the velocity.

[0060] Initialize external archive A= (Empty set) After generating the initial particle swarm, calculate the multi-objective fitness value (such as path length, energy consumption, and safety) of each particle, and filter the initial non-dominated solutions through non-dominated sorting and store them. The archive maintenance mechanism (such as dominance relationship judgment, crowding distance calculation, and truncation strategy) is dynamically updated in subsequent iterations to ensure that it stores the current optimal and diverse solution set.

[0061] Step 3 Fitness Evaluation: Based on the constructed digital map and multi-objective optimization requirements (path length, energy consumption, safety), calculate the fitness value of the path scheme represented by each particle in the current population;

[0062] The fitness value expression is: ,in, The fitness value is used to measure the quality of the path scheme represented by each particle in the current population. It represents the distance from the starting point to the ending point; This represents the energy consumed by the proposed path. Indicators representing the safety of the route plan. Indicates the time required to complete the route plan. , , , The represents the weighting coefficient, used to balance the relationship between the four objectives of path length, energy consumption, safety, and time.

[0063] Step 4: Update the external archive: Add non-dominated solutions from the current population that are not dominated by any archived solutions to the archive, and remove old solutions dominated by newly added solutions; if the archive size exceeds the preset capacity, use the crowding distance sorting strategy to eliminate inferior solutions in the "crowded area" to maintain the diversity and distribution of archived solutions;

[0064] Filtering newly added non-dominated solutions: For each particle in the current population, calculate its dominance relationship with all solutions in the external archive. If the particle is not dominated by any solution in the archive, add it to the external archive. The specific process is as follows:

[0065] For each solution q in the current particle and the external archive A, determine whether q dominates the current particle (i.e., q is not inferior to the current particle on all targets and is strictly superior to p on at least one target); if it is not dominated by any solution in the archive, add it to the external archive; otherwise, do not add it; after adding a new solution to the external archive, old solutions dominated by the newly added solution need to be removed; for each solution in the archive, determine whether there is a newly added solution that dominates it; if so, remove it from the archive; when the capacity of the external archive reaches the maximum limit, a crowding sorting strategy is used to eliminate inferior solutions in the "crowded region" to ensure the diversity of solution distribution within the archive;

[0066] For each solution, calculate its crowding degree in the target space; the formula for calculating the crowding degree is: ,

[0067] in, ‎ represents the solution The degree of congestion is used to measure the degree of congestion of the solution. The relative density in the target space; Representing the solution Its adjacent solutions In the The difference in values ​​on the objective function reflects the relative distance between the two on that objective; This indicates the number of objective functions, i.e., how many optimization objectives need to be considered; Representing the solution In the The values ​​of each objective function; Representing the solution Adjacent solutions The value of the i-th objective function;

[0068] Sort the solutions in the archive in descending order according to the crowding degree (crowding distance), retain the N solutions with the highest crowding degree (the larger the crowding distance, the sparser the solution distribution in the target space), and discard the rest of the solutions.

[0069] Step 5: Particle velocity and position update: Update the velocity of each particle according to Subgroup Optimization (PSO) based on the individual historical optimal solution of the particle.

[0070] The speed update expression is: ,in, Represents particles In dimensions The speed on, Represents particles In dimensions Current position on Represents particles In dimensions The best historical position on Represents all particles in dimension The global optimal position on , A random number between [0,1] is used to increase the randomness and diversity of the algorithm; Inertial weights are used to balance the influence of a particle's historical velocity on its current velocity. and These are cognitive learning factors and social learning factors, which respectively regulate the degree to which particles learn towards individual and global optima.

[0071] Step 6 Adaptive Mutation Operation: Based on the current iteration number and particle fitness value, dynamically adjust the mutation probability and mutation magnitude, and perform mutation operations on some particles to increase population diversity and avoid getting trapped in local optima;

[0072] The mutation probability decreases as the number of iterations increases, and the specific calculation formula is as follows: ,in, The mutation probability is the number of iterations. The initial mutation probability, This represents the current iteration number. The maximum number of iterations,

[0073] The mutation amplitude decreases as the particle fitness value increases, and the specific calculation formula is as follows: ,in, This represents the variation range of the current particle. The initial variation amplitude, This represents the current particle's fitness value. This represents the maximum fitness value in the current population.

[0074] For each particle, generate a random number r. If r < 0, then... Then, a mutation operation is performed on the particle, the specific formula of which is: ,in, For the first The first particle Dimensional position, These are random numbers that follow a standard normal distribution.

[0075] Step 7: Iteration Termination Judgment: If the termination condition is met, the algorithm ends and the optimal solution in the external archive is output as the final result; otherwise, return to step 3 to continue execution.

[0076] Iteration termination judgment: If the maximum number of iterations T is reached or the fitness value has not significantly improved for N consecutive iterations, exit the loop and proceed to step 8; otherwise, return to step 3; select the non-dominated solution with the highest comprehensive score from the external archive as the final optimal path.

[0077] Step 8: Optimal Path Generation and Command Conversion: Select the optimal path scheme from the final results and convert it into flight commands that the UAV can execute;

[0078] Optimal path selection involves normalizing path length, energy consumption, and safety from multi-objective solutions (e.g., using the range method to eliminate dimensional differences) and assigning weights based on task priority (e.g., reconnaissance missions prioritize safety, while transportation missions prioritize energy consumption).

[0079] Calculate the overall score for each non-dominated solution using the weighted sum method: ,in, For the target weight, The target value is the normalized value; the solution with the highest comprehensive score is selected as the final optimal path (e.g., the path with the highest score also satisfies "short length, low energy consumption, and safe obstacle avoidance").

[0080] Flight command conversion (from path to executable command): Path smoothing: Cubic spline interpolation is used to smooth the path points to ensure path continuity; the smoothed path is checked to see if it meets the UAV speed limits (e.g., maximum flight speed ≤ 20 m / s), acceleration limits (e.g., maximum acceleration ≤ 5 m / s²), and endurance constraints (e.g., total flight time ≤ battery life). If not, the path is readjusted; if it meets the requirements, the smoothed path is discretized into a sequence of waypoints at fixed intervals (e.g., one point per meter). Each waypoint contains coordinates, velocity, and attitude information for UAV flight control. The waypoint sequence is packaged into a command packet using a communication protocol and sent to the UAV flight control system via a wireless link.

[0081] Application scenario: When three drones are needed to transport supplies from an emergency warehouse (coordinates 116.3 degrees east longitude, 39.9 degrees north latitude) to a designated location. The flight area is a 2 km × 2 km urban area, including high-rise building clusters (obstacle coverage 60%), with a no-fly zone within 300 meters of the hospital.

[0082] Environmental modeling

[0083] Dynamic grid division: The flight area is divided into a basic grid of 20m x 20m (100 x 100 grids in total), and the grid is further densified to 10.5m x 10.5m around the hospital no-fly zone. Buildings exceeding 15m in height are detected by lidar and marked as impassable areas.

[0084] When a temporary obstacle (such as a broken-down rescue vehicle) is added, the safe path boundary within a 30-meter radius is recalculated based on the center point of the obstacle. For example, the equation of the path boundary between the rescue vehicle's coordinates (116.312 degrees east longitude, 39.908 degrees north latitude) and the adjacent building is: 50 times the east longitude coordinates plus 50 times the north latitude coordinates equals 106250.

[0085] Path scheme initialization

[0086] Population setup: Generate 100 initial path schemes, each containing the flight paths of 3 drones (each drone's path has 4 waypoints, including the origin and destination). If a waypoint falls into a building area, the coordinates of that point will be randomly adjusted within 5% of that point's coordinates.

[0087] Optimize target weights: path length accounts for 30%, energy consumption accounts for 40%, and safety accounts for 30% (prioritizing battery life and obstacle avoidance capabilities).

[0088] Solution Iteration and Optimization

[0089] Adaptive adjustment: In 500 iterations, the path mutation probability is maintained at 9% for the first 250 iterations, and then gradually reduced to 4% in the later iterations. The mutation range changes dynamically with the merits of the proposed solution, and the mutation range of the excellent solution is less than 37 meters.

[0090] Optimal solution selection: The final solution with a comprehensive score of 0.41 (path length score 0.20, energy consumption score 0.15, and safety score 0.90) was selected, which is significantly better than the traditional algorithm by 18%.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive UAV swarm path planning method, characterized in that, Includes the following steps: Step 1: Flight Environment Modeling. After receiving the path planning task trigger signal, the first step is to model the UAV's flight environment. After modeling, the environmental information is converted into a computer-readable digital map. Step 2: Algorithm parameter setting and particle initialization: After completing the environment modeling, set the parameters of the particle swarm optimization algorithm. Then, randomly initialize the position and velocity of the particles. At the same time, initialize the external archive to an empty set to store non-dominated solutions. Step 3 Fitness Evaluation: Based on the constructed digital map and multi-objective optimization requirements, calculate the fitness value of the path scheme represented by each particle in the current population; Step 4: Update the external archive: Add non-dominated solutions from the current population that are not dominated by any archived solution to the archive, and remove old solutions dominated by newly added solutions; if the archive size exceeds the preset capacity, use the crowding distance sorting strategy to eliminate inferior solutions in the "crowded region". Step 5: Particle velocity and position update: Based on the individual historical optimal solution of each particle, update the velocity of each particle according to subgroup optimization; Step 6: Adaptive Mutation Operation: Based on the current iteration number and particle fitness value, dynamically adjust the mutation probability and mutation magnitude, and perform mutation operations on some particles. Step 7: Iteration Termination Judgment: If the termination condition is met, the algorithm ends and the optimal solution in the external archive is output as the final result; Otherwise, return to step 3 and continue execution; Step 8: Optimal Path Generation and Command Conversion: Select the optimal path scheme from the final results and convert it into flight commands that the UAV can execute.

2. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The modeling method adopts the grid method, which divides the flight area into uniform regular grids, and the resolution is dynamically adjusted according to mission requirements; the formula is: grid side length s=L / N, where L is the side length of the area and N is the number of grids; Obstacle point cloud data is acquired using LiDAR and visual sensors. If an obstacle point cloud exists within the grid, it is marked as an "obstacle grid"; otherwise, it is marked as a "passable grid".

3. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The modeling method employs the Voronoi diagram method, which uses obstacle vertices or center points as generators to ensure coverage of all obstacles. It generates equidistant lines of the Voronoi boundary point-by-point using scan lines, where Pi(…)… , ) and Pj( , The equation of the perpendicular bisector of () is: The drone's path is within the Voronoi region, maintaining a set safe distance from obstacles; when new obstacles are added, the Voronoi map of the affected area is recalculated. Digital maps collect three-dimensional data of the flight environment through sensors, including the location, shape, and size of obstacles; the collected data is preprocessed, including noise reduction, smoothing, and segmentation operations; and the processed data is converted into a digital map according to the selected modeling method; the generated digital map is stored in a computer database, which supports subsequent algorithm calls.

4. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The parameters of the algorithm include, but are not limited to, population size, maximum number of iterations, inertia weight, and learning factor; each particle corresponds to a path planning scheme for a drone swarm, transforming the path problem into a numerical encoding of decision variables; Particle position initialization: First, assume there are M drones in the cluster, and each drone's path consists of K nodes. Then, the dimension of a single particle is D = M × K × 2. The particle position is randomly generated within the path coordinate boundaries, expressed as: ,in, Indicates the first The particle in the first The initial position in each dimension, where Represents the particle number. Represents the dimension number; Indicates the first Minimum boundary value in each dimension Indicates the first Maximum boundary value in each dimension This represents a random number uniformly distributed in the interval [0,1], used to generate random positions for particles; If the initial position corresponds to a drone path that traverses obstacles, adjustments are necessary. These adjustments can be made by randomly perturbing and fine-tuning the point coordinates or by regenerating the values ​​in that dimension, ensuring that the adjusted initial solution satisfies the basic constraints of the drone's flight. The velocity determines the particle's search direction and step size. The initial velocity needs to be small to avoid jumping out of the feasible region. The expression for initializing the particle velocity is: ,in, Indicates the first The particle in the first Initial velocity in each dimension Indicates the first The particle in the first The minimum velocity in the dimension, Indicates the first The particle in the first The maximum value of the velocity in the dimension, This represents a random number uniformly distributed in the interval [0,1], used to randomly generate an initial velocity between the minimum and maximum values, ensuring the randomness and diversity of the velocity. Initialize external archive A= After generating the initial particle swarm, the multi-objective fitness value of each particle is calculated, and the initial non-dominated solutions are selected and stored through non-dominated sorting. The archive maintenance mechanism is dynamically updated in subsequent iterations to ensure that it stores the current optimal and diverse set of solutions.

5. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The fitness value expression is: ,in, The fitness value is used to measure the quality of the path scheme represented by each particle in the current population. It represents the distance from the starting point to the ending point; This represents the energy consumed by the proposed path. Indicators representing the safety of the route plan. This indicates the time required to complete the route plan. , , , The represents the weighting coefficient, used to balance the relationship between the four objectives of path length, energy consumption, safety, and time.

6. The adaptive UAV swarm path planning method according to claim 1, characterized in that, Filtering newly added non-dominated solutions: For each particle in the current population, calculate its dominance relationship with all solutions in the external archive. If the particle is not dominated by any solution in the archive, add it to the external archive. The specific process is as follows: For each solution q in the current particle and the external archive A, determine whether q dominates the current particle; if it is not dominated by any solution in the archive, add it to the external archive; otherwise, do not add it. After adding a new solution to the external archive, old solutions dominated by the newly added solution need to be removed. For each solution in the archive, determine whether there is a newly added solution that dominates it; if so, remove it from the archive. When the capacity of the external archive reaches its maximum limit, a crowding sorting strategy is used to eliminate inferior solutions in "crowded regions" to ensure the diversity of solution distribution within the archive. For each solution, calculate its crowding degree in the target space; the formula for calculating the crowding degree is: , in, ‎ represents the solution The degree of congestion is used to measure the degree of congestion of the solution. The relative density in the target space; Representing the solution Its adjacent solutions In the The difference in values ​​on each objective function; This indicates the number of objective functions, i.e., how many optimization objectives need to be considered; Representing the solution In the The values ​​of each objective function; Representing the solution Adjacent solutions The value of the i-th objective function; Sort the solutions in the archive in descending order according to their crowding level, retain the N solutions with the highest crowding level, and discard the rest.

7. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The speed update expression is: ,in, Represents particles In dimensions The speed on, Represents particles In dimensions Current position on Represents particles In dimensions The best historical position on Represents all particles in dimension The global optimal position on , A random number between [0,1] is used to increase the randomness and diversity of the algorithm; Inertial weights are used to balance the influence of a particle's historical velocity on its current velocity. and These are cognitive learning factors and social learning factors, which respectively regulate the degree to which particles learn towards individual and global optima.

8. The adaptive UAV swarm path planning method according to claim 1, characterized in that, The mutation probability decreases as the number of iterations increases, and the specific calculation formula is as follows: ,in, The mutation probability is the number of iterations. The initial mutation probability, This represents the current iteration number. The maximum number of iterations, The mutation amplitude decreases as the particle fitness value increases, and the specific calculation formula is as follows: ,in, This represents the variation range of the current particle. The initial variation amplitude, This represents the current particle's fitness value. This represents the maximum fitness value in the current population. For each particle, generate a random number r. If r < 0, then... Then, a mutation operation is performed on the particle, the specific formula of which is: ,in, For the first The first particle Dimensional position, These are random numbers that follow a standard normal distribution.

9. The adaptive UAV swarm path planning method according to claim 1, characterized in that, Iteration termination judgment: If the maximum number of iterations T is reached or the fitness value has not significantly improved for N consecutive iterations, exit the loop and proceed to step 8; otherwise, return to step 3; select the non-dominated solution with the highest comprehensive score from the external archive as the final optimal path.

10. The adaptive UAV swarm path planning method according to claim 1, characterized in that, Optimal path selection involves normalizing path length, energy consumption, and safety from multi-objective solutions and assigning weights based on task priority. Calculate the overall score for each non-dominated solution using the weighted sum method: ,in, For the target weight, The normalized target value is used; the solution with the highest comprehensive score is selected as the final optimal path. Flight command conversion path smoothing: Cubic spline interpolation is used to smooth the path points to ensure path continuity; the smoothed path is checked to see if it meets the speed limit, acceleration limit and endurance constraint of the UAV. If it does not meet the constraint, the path is readjusted; if it meets the constraint, the smoothed path is discretized into a sequence of waypoints at fixed intervals. Each waypoint contains coordinate, speed and attitude information for UAV flight control. The waypoint sequence is packaged into a command packet using a communication protocol and sent to the UAV flight control system via a wireless link.

Images