Unmanned aerial vehicle optimal path adaptive planning method using particle swarm optimization

Abstract

The unmanned aerial vehicle optimal path adaptive planning method based on particle swarm optimization relates to the field of unmanned aerial vehicle path planning, and comprises the following steps: constructing a spatial constraint model of an unmanned aerial vehicle flight task; initializing a particle swarm based on the spatial constraint model; combining path smoothness, path threat probability and voyage efficiency to establish a multi-objective fitness function, and calculating the fitness value of each particle based on the particle swarm; performing hierarchical particle swarm optimization iteration, in each iteration, adaptively adjusting the control parameters based on the current particle swarm distribution characteristics, and updating the path node position of the particles; when the convergence condition is met, outputting the global optimal path as the final flight trajectory of the unmanned aerial vehicle, and controlling the unmanned aerial vehicle to execute. The unmanned aerial vehicle optimal path adaptive planning method based on particle swarm optimization solves the problems that the unmanned aerial vehicle route path optimization target is single and difficult to coordinate, and the route planning path effect is poor.

Description

Adaptive Path Planning Method for Unmanned Aerial Vehicles Using Particle Swarm Optimization Technical Field

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

[0002] The Particle Swarm Optimization (PSO) algorithm starts with a random solution, iteratively searches for the currently found optimal value to find the global optimum, and evaluates the solution by fitness.

[0003] Existing route planning methods suffer from poor optimal path performance, slow convergence speed, and susceptibility to local optima. This results in unmanned aerial vehicle (UAV) route optimization having a singular objective and difficulty in coordination, leading to unsatisfactory route planning results. Therefore, a particle swarm optimization method is needed to achieve real-time, efficient, and intelligent adaptive optimal path planning for UAVs. Summary of the Invention

[0004] This application employs a particle swarm optimization-based adaptive planning method for UAV optimal paths, aiming to address the problem in existing technologies where UAV route path optimization has a single objective and is difficult to coordinate, resulting in poor route planning performance.

[0005] In view of the above problems, this application adopts a particle swarm optimization-based adaptive path planning method for UAVs.

[0006] This application employs a particle swarm optimization-based adaptive path planning method for unmanned aerial vehicles, including:

[0007] Construct a spatial constraint model for UAV flight missions, wherein the spatial constraint model includes an obstacle distribution field and flight corridor boundaries;

[0008] The particle swarm is initialized based on the spatial constraint model, wherein each particle represents a candidate path, and the control parameters of the particle swarm include path smoothness weight, threat avoidance weight, and range cost weight.

[0009] A multi-objective fitness function is established by combining path smoothness, path threat probability, and range efficiency, and the fitness value of each particle is calculated based on the particle swarm.

[0010] Perform hierarchical particle swarm optimization iterations. In each iteration, the control parameters are adaptively adjusted based on the current particle swarm distribution characteristics, and the path node positions of the particles are updated.

[0011] When the convergence condition is met, the globally optimal path is output as the final flight trajectory of the UAV, and the UAV is controlled to execute it.

[0012] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0013] This application employs a particle swarm optimization (PSO) method for UAV path planning. By adaptively adjusting parameters, it generates optimal flight paths in complex environments. Through spatial constraint modeling, multi-objective optimization, and hierarchical iteration, it improves path safety, smoothness, and efficiency, realizing a spatial constraint model for environmental threats and enhancing practicality and reliability. An adaptive adjustment mechanism based on path characteristics adaptively reduces path threat probabilities, establishing an intelligent evaluation system capable of dynamically balancing multiple objectives. A hierarchical adaptive optimization strategy performs staged, adaptive parameter adjustments for more efficient global optimal path finding. A path feasibility repair mechanism ensures the continuity and stability of the optimization process, while an iterative convergence judgment mechanism intelligently determines the convergence state and dynamically adjusts the path. This solves the problems of single and difficult-to-coordinate optimization objectives and poor path planning performance in UAV path planning. It ensures the stability of the output of the global optimal path, providing an adaptive and executable optimal UAV path. It achieves high-precision environmental modeling, multi-objective adaptive trade-offs, and intelligent search optimization for UAV path planning. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0015] Figure 1 is a flowchart illustrating the adaptive path planning method for UAVs using particle swarm optimization.

[0016] Figure 2 is a schematic diagram of the process of performing hierarchical particle swarm optimization iteration and updating the path node positions of particles in the adaptive planning method for the optimal path of UAVs using particle swarm optimization. Detailed Implementation

[0017] This application employs a particle swarm optimization-based adaptive planning method for UAV optimal paths to address the problems of poor performance, slow convergence speed, and susceptibility to local optima in existing technologies, which result in poor performance of UAV flight path planning.

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0020] Example 1, as shown in Figure 1, this application employs a particle swarm optimization-based adaptive path planning method for unmanned aerial vehicles (UAVs). The method includes:

[0021] S10: Construct a spatial constraint model for the UAV flight mission, wherein the spatial constraint model includes an obstacle distribution field and a flight corridor boundary;

[0022] In this embodiment of the application, a constraint model for UAV flight mission is constructed. The spatial constraint model includes an obstacle distribution field and a flight corridor boundary. The obstacle distribution field is the regional distribution field of the danger zone and the obstacle zone. The flight corridor boundary is the channel range in which the UAV can fly safely.

[0023] Step S10 in the method provided in this application embodiment includes:

[0024] Acquire multi-source terrain data of the target task area and construct an obstacle probability field based on the multi-source terrain data, wherein the multi-source terrain data includes a digital elevation model, building distribution data and no-fly zone coordinates;

[0025] Based on the mission start point, mission end point, and the obstacle probability field, a fast exploration random tree algorithm is used to generate the flight corridor boundary.

[0026] Based on the aforementioned obstacle probability field and flight corridor boundary, a spatial constraint model is established that includes static obstacle constraints and dynamic airspace constraints.

[0027] Specifically, multi-source terrain data of the target mission area is acquired, and an obstacle probability field is constructed based on this data. The multi-source terrain data includes a digital elevation model (DEM), building distribution data, and no-fly zone coordinates. The terrain data of the target area is recorded digitally, and the ground is divided into a grid to obtain multi-terrain data. The obstacle probability field is then constructed using the DEM, building distribution data, and no-fly zone coordinates from this multi-terrain data. This obstacle probability field represents the threat probability at different locations.

[0028] For example, in mountainous tasks, areas with a slope greater than 25 degrees are identified as danger zones, and a digital elevation model showing elevation changes is obtained; building data is obtained through city maps; the coordinates of the no-fly zone are the coordinates of a 5-kilometer radius around the airport.

[0029] Furthermore, based on the task start point, task end point, and obstacle probability field, a fast search random tree algorithm is used to generate the flight corridor boundary. The fast search random tree (RRT) is an incremental algorithm for path planning applied in the field of robot motion planning. It is a method that rapidly constructs a tree structure in free space through random sampling to explore possible paths. Inputting the task start point, task end point, and obstacle probability field into the fast search random tree yields the corresponding flight corridor boundary. The flight corridor boundary is the safe area connecting the start and end points.

[0030] For example, the mission start point, mission end point, and obstacle probability field are input into a fast search random tree, and a three-dimensional flight corridor boundary is constructed using multi-source terrain data, which is a passage 200 meters wide and 150 meters high.

[0031] Furthermore, by integrating the obstacle probability field and the flight corridor boundary, a spatial constraint model incorporating static obstacle constraints and dynamic airspace constraints is established. First, the data is initialized. Then, forward reasoning is used to limit the search range for address pattern extraction. Spatial constraint relationships are used for reasoning and recognition, transforming the obstacle probability field and flight corridor boundary into spatial constraint address representation patterns. Finally, these patterns are matched with an address database to obtain the final target address. Static obstacle constraints are based on the spatial constraint model constructed from the ground obstacle probability field, while dynamic airspace constraints are based on the spatial constraint model formed by the flight corridor boundary.

[0032] Step S10 in this embodiment of the application, which involves constructing an obstacle probability field from multi-source terrain data, includes:

[0033] The digital elevation model is subjected to terrain gradient analysis to identify areas with slopes greater than a preset slope threshold as terrain obstacle areas.

[0034] Perform 3D convex hull modeling on the building distribution data and calculate the influence range of each building;

[0035] Based on the coordinates of the no-fly zone, establish a polygonal constraint boundary to determine the absolute no-fly zone;

[0036] Different safe distance thresholds are set based on the terrain obstacle areas, the influence range of buildings, and the absolute no-fly zones.

[0037] An obstacle threat probability distribution model is established using an exponential decay function, where the threat probability in the center region of the obstacle is 1, and decreases to 0 with a preset decay coefficient as the safe distance increases.

[0038] By integrating the threat probability distributions of the terrain obstacle areas, the influence range of buildings, and the absolute no-fly zones, a unified obstacle probability field is generated.

[0039] Specifically, topographic data of the target area is recorded digitally, the ground is divided into grids to obtain topographic data for multiple regions, and topographic gradient analysis is performed on the digital elevation model (DEM). Regions with slopes greater than a preset slope threshold are identified as topographic obstacle zones. The DEM is a digital simulation of the ground topography using limited topographic elevation data; it is a physical ground model representing ground elevation using an ordered numerical array. Topographic gradient analysis is performed on the DEM based on slope grading data. The gradient levels are as follows: 0–2° is level 1 (flat ground), 2–6° is level 2 (gentle slope), 6–15° is level 3 (gentle slope), 15–25° is level 4 (sloping slope), and above 25° is level 5 (steep slope). Multiple regions in the DEM are identified based on the topographic gradient analysis. The identified slopes are compared with preset slope thresholds, and regions with slopes greater than the preset thresholds are selected as topographic obstacle zones. The preset slope thresholds can be dynamically set to adapt to the needs of topographic gradient analysis.

[0040] For example, the area is divided into 4 regions using a grid. The preset slope threshold is 15°. The slopes are identified as 10°, 20°, 13°, and 18°. The regions with slopes of 20° and 18° are selected as terrain obstacle areas.

[0041] Furthermore, a three-dimensional convex hull counting algorithm is used to perform three-dimensional convex hull modeling on the building distribution data, constructing the smallest convex polyhedron that can completely enclose the buildings, and then using the convex hull to accelerate data calculation to obtain the influence range of each building.

[0042] For example, after modeling a 3D convex hull of a building, the influence range extends outward by 50 meters.

[0043] Furthermore, based on the geographical coordinates of the no-fly zone, polygonal constraint boundaries, such as rectangles or circles, are established within the geographical area, and the area inside the established polygonal constraint boundaries is defined as an absolute no-fly zone.

[0044] For example, a rectangular area is determined as the absolute no-fly zone coordinates based on the plane coordinates of the no-fly zone (75°, 105°), (90°, 130°), (75°, 130°), (90°, 105°).

[0045] Furthermore, different obstacle types correspond to different danger zones and safe distances. Based on terrain obstacle zones, building influence zones, and absolute no-fly zones, different safe distance thresholds are set for different types of obstacles.

[0046] For example, the safe distance for terrain obstacle zones is 20 meters, for buildings it is 50 meters, and for absolute no-fly zones it is 100 meters.

[0047] Furthermore, since the threat probability in the center area of ​​an obstacle decreases at a rate proportional to the current value with distance, exhibiting an exponential decay, an obstacle threat probability distribution model is established using an exponential decay function. In this model, the center area of ​​the obstacle is the most dangerous, with a threat probability of 1. As the safe distance increases, the probability decreases to 0 according to a preset decay coefficient. When the threat probability is 0, the area is completely safe, with a decay coefficient of 0.1 / m.

[0048] The danger level drops to 0 when you are far enough away, which is equivalent to being completely safe.

[0049] For example, on a mountain at an altitude of 500 meters, slope analysis shows that the slope within 100 meters of the summit exceeds 25 degrees, with a threat probability of 1.0. Using an exponential decay function, the threat probability decreases by 0.1 per meter beyond a safe distance of 20 meters.

[0050] Furthermore, by using images to fuse the threat probability distribution data of terrain obstacle areas, building influence ranges, and absolute no-fly zones, a unified obstacle probability field is generated, which includes dangerous areas, danger influence ranges, and threat probability distributions. Through the obstacle probability field, a certain degree of danger avoidance can be achieved, ensuring flight route safety.

[0051] For example, terrain obstacle areas with slopes greater than the preset recognition threshold of 20° and 18°, buildings should be extended 50 meters outward from the range, and the distribution of safety distances should be fused to obtain a unified obstacle probability field.

[0052] In this embodiment, by integrating multi-source terrain data such as digital elevation model, building distribution data, and no-fly zone coordinates, and using an exponential decay function to establish an obstacle probability field, the flight corridor boundary is generated. This unifies static obstacle avoidance and dynamic airspace management constraints into a spatial constraint model, ensuring the feasibility of the planned path and greatly improving the practicality and reliability of the method.

[0053] S20: Initialize the particle swarm based on the spatial constraint model, wherein each particle represents a candidate path, and the control parameters of the particle swarm include path smoothness weight, threat avoidance weight, and range cost weight.

[0054] In this embodiment of the application, a large number of possible flight paths are generated through a spatial constraint model, where each particle is a candidate path, and all possible paths form a particle swarm. The control parameters of the particle swarm include path smoothness weight, threat avoidance weight, and range cost weight.

[0055] Step S20 in the method provided in this application embodiment includes:

[0056] Based on the coordinates of the mission start and end points, an initial set of path nodes is randomly generated within the boundary of the flight corridor using Logistic chaotic mapping.

[0057] Based on the obstacle probability field, the initial path node set is screened for feasibility, and path nodes with threat probabilities exceeding a preset threat probability threshold are removed.

[0058] The initial values ​​of the control parameters for the particle swarm are set, wherein the path smoothness weight is determined based on the curvature variance of the initial path nodes, the threat avoidance weight is determined based on the maximum threat value of the obstacle probability field, and the range cost weight is determined based on the Euclidean distance from the mission start point to the mission end point.

[0059] After filtering, the path nodes that meet the requirements are arranged in order from the start point of the task to the end point of the task to form a path node sequence. Each path node sequence corresponds to a particle, and multiple particles together form a particle swarm.

[0060] Specifically, based on the coordinates of the task's starting and ending points, an initial set of path nodes is randomly generated within the flight corridor boundary using a Logistic chaotic mapping. The Logistic chaotic mapping is a random number generation method that produces relatively uniform data distribution, avoiding data clustering and duplication. Furthermore, as a highly nonlinear and sensitive dynamic system to initial conditions, its trajectory exhibits randomness and unpredictability in space. Therefore, chaotic mapping possesses unique advantages in optimization problems; adjusting its parameters can significantly affect the performance of the optimization algorithm. Properly designed mapping parameters can improve convergence speed and global search capability, effectively avoiding local optima and improving search efficiency. Using a simulated random method to randomly generate the path node sequence within the flight corridor achieves comprehensive coverage and ensures diversity.

[0061] For example, a sequence of path nodes for 100 particles is generated using a Logistic chaotic mapping from the starting point (0, 0, 0) to the ending point (1000, 1000, 100).

[0062] Furthermore, the initial path node set is subjected to feasibility screening based on the obstacle probability field, and path nodes with threat probabilities exceeding a preset threat probability threshold are removed. This preset threat threshold is derived from the drone's maximum risk resistance capability. If the threat probability exceeds the preset threshold, it indicates that the path node is too close to the center of the obstacle, and the security risk exceeds the drone's risk resistance capability; therefore, it must be removed.

[0063] For example, the preset threat probability threshold is 0.8, and each particle has 10 nodes. After filtering, nodes that fall on buildings with a threat probability of 0.8 or higher are removed.

[0064] Furthermore, based on environmental dynamics, initial values ​​for the particle swarm control parameters are set. The path smoothness weight is determined based on the curvature variance of the initial path nodes, the threat avoidance weight is determined based on the maximum threat value of the obstacle probability field, and the range cost weight is determined based on the Euclidean distance from the mission start point to the mission end point. Path smoothness measures the stability of the UAV's flight, threat avoidance measures the degree to which the UAV avoids hazards, and range cost measures the distance between flight paths. By setting the path smoothness weight, threat avoidance weight, and range cost weight, the initial values ​​for the particle swarm control parameters are obtained, and the particle swarm is controlled.

[0065] For example, the initial values ​​of the control parameters are set as follows: smoothness weight 0.3, threat avoidance weight 0.4, and range cost weight 0.2.

[0066] Furthermore, the selected path nodes that meet the requirements are arranged in order from the start point of the task to the end point of the task to construct a path node sequence. Each path node sequence corresponds to a particle, and multiple corresponding particles together form a particle swarm.

[0067] For example, a particle swarm containing 5 particles is obtained by following the path node sequence of 5 path nodes corresponding to 5 paths.

[0068] In this embodiment, a sequence of path nodes is randomly generated within the flight corridor using a Logistic chaotic mapping to ensure diversity. Then, based on the obstacle probability field, feasibility screening is performed to remove nodes whose threat probability exceeds a threshold. Finally, initial values ​​for the particle swarm control parameters are set based on environmental dynamics. By using chaotic mapping to improve convergence speed and global search capability, local optima are effectively avoided, search efficiency is improved, and comprehensive coverage is achieved, ensuring diversity.

[0069] S30: Combine path smoothness, path threat probability and range efficiency to establish a multi-objective fitness function, and calculate the fitness value of each particle based on the particle swarm.

[0070] In this embodiment, path smoothness, path threat probability, and flight efficiency are integrated into a unified standard that can score each path through a multi-objective fitness function, thus balancing multiple objectives such as path smoothness, path threat probability, and flight efficiency.

[0071] Step S30 in the method provided in this application embodiment includes:

[0072] The curvature change of the path node sequence is analyzed, and the average curvature change is calculated based on the rate of change of the angle between the vectors formed by three consecutive path nodes, which is used as the path smoothness.

[0073] The maximum probability value of each path node in the obstacle probability field is extracted as the path threat probability. Based on the path smoothness, the path threat probability is adaptively reduced to obtain the corrected path threat probability.

[0074] The path range efficiency is obtained by calculating the ratio of the total path length to the Euclidean distance from the mission start point to the mission end point.

[0075] A multi-objective fitness function is constructed by combining the modified path threat probability, path smoothness, and path range efficiency using a weighted summation method.

[0076] The fitness value of each particle in the particle swarm is calculated based on the multi-objective fitness function.

[0077] Specifically, the curvature changes of the path node sequence are analyzed, and the average curvature change is calculated based on the rate of change of the angle between the vectors formed by three consecutive path nodes, which serves as the path smoothness. Curvature is the rate of rotation of the tangent direction angle at a point on a curve with respect to the arc length. It indicates the degree to which the curve deviates from a straight line. The greater the curvature, the greater the curvature, and the greater the average curvature change. This means the UAV bends more on the path, and the lower the path smoothness. The rate of change of the vector angle is the inverse cosine function of the ratio of the vector product to the product of the vector magnitudes, divided by time. The average rate of change of the vector angle between three consecutive path nodes is calculated as the path smoothness. The smoother the path, and the lower the path smoothness, the greater the likelihood that the path will be selected as the optimal flight path.

[0078] For example, for three consecutive path nodes a, b, and c, the rate of change of the vector angle at each node is calculated to be 0.12 rad / s, 0.15 rad / s, and 0.18 rad / s, respectively. The average curvature change is [(0.12+0.15+0.18) / 3]=0.15 rad / s, and the path smoothness is 0.15.

[0079] Furthermore, the maximum probability value of each path node in the obstacle probability field is extracted as the path threat probability. Based on path smoothness, the path threat probability is adaptively reduced to obtain the corrected path threat probability. The safety of the entire path is determined by the most dangerous point; therefore, the maximum probability value of each path node in the obstacle probability field is selected as the path threat probability. An exponential function is used to reduce the path threat probability using path smoothness, resulting in a reduced path threat probability. A higher corrected path threat probability indicates a lower path safety and minimizes the likelihood of it being selected as the optimal path.

[0080] Furthermore, the path range efficiency is obtained by calculating the ratio of the total path length to the Euclidean distance from the mission start point to the mission end point. Path range efficiency = Total path length / Euclidean distance from mission start point to mission end point, where the total path length is 1000 meters, and the Euclidean distance is the shortest straight-line distance between two points, which can be calculated using the Pythagorean theorem. The smaller the Euclidean distance from the mission start point to the mission end point, the greater the path range efficiency; the closer the ratio is to 1, the higher the path range efficiency. Path range efficiency reflects the length of the path; the greater the path range efficiency, the more likely the path is to be designated as the optimal route.

[0081] For example, the mission start point is (0, 0, 0), and the mission end point is (1000, 1000, 100), with a Euclidean distance of approximately 1417 meters. The path range efficiency is 1000 / 1417 = 70.5%.

[0082] Furthermore, a weighted summation method is used to combine the modified path threat probability, path smoothness, and path range efficiency to construct a multi-objective fitness function, which can be used for the optimal selection of more than one objective in a given area. Specifically, the modified path threat probability is negatively correlated with fitness, while path smoothness and path range efficiency are positively correlated with fitness.

[0083] Furthermore, based on the multi-objective fitness function, the fitness value of each particle in the particle swarm is calculated. Since the multi-objective fitness is the sum of the products of 1 - the modified path threat probability, path smoothness, and path range efficiency with their respective weights, the calculation formula is: Multi-objective fitness = (1 - modified path threat probability) × w1 + Path smoothness × w2 + Path range efficiency × w3. The smaller the fitness value, the higher the probability of selecting the path.

[0084] For example, if the weights for the modified path threat probability, path smoothness, and path range efficiency are set to 0.3, 0.4, and 0.2 respectively, the multi-target fitness is calculated as (1-0.435)×0.3+0.15×0.4+70.5%×0.2≈0.37.

[0085] Step S30 of the method provided in this application embodiment adaptively reduces the path threat probability based on the path smoothness to obtain a corrected path threat probability, including:

[0086] The threat probability reduction coefficient is calculated based on the path smoothness, wherein the threat probability reduction coefficient and the path smoothness satisfy a negatively correlated exponential function relationship;

[0087] The path threat probability is corrected by using the threat probability reduction factor to obtain the corrected path threat probability.

[0088] Specifically, the threat probability reduction coefficient is calculated based on path smoothness, where the threat probability reduction coefficient and path smoothness satisfy a negatively correlated exponential function relationship. Threat probability reduction coefficient = e( -k×路径平滑度） Where e is a natural constant, approximately 2.718, and k is a preset positive coefficient used to adjust the rate of change of the reduction coefficient with path smoothness. It is determined according to the actual flight scenario, and the resulting threat probability reduction coefficient is a number between 0 and 1.

[0089] For example, with a path smoothness of 0.16, a maximum threat probability of 0.6, and K = 2, the threat probability reduction factor is 2.718. （-2×0.16） =0.726.

[0090] Furthermore, the path threat probability is corrected using a threat probability reduction factor to obtain the corrected path threat probability. Corrected path threat probability = Original path threat probability × Threat probability reduction factor. The corrected path threat probability allows the assessment of path risk to better reflect actual physical conditions, rather than relying solely on static threats from obstacles.

[0091] For example, the path smoothness is 0.16, the maximum threat probability is 0.6, and the corrected path threat probability is 0.6 × 0.726 ≈ 0.435.

[0092] In this embodiment, a fixed-weighted linear weighting method is used to construct an objective function by combining path smoothness, modified path threat probability, and path range efficiency. This function is dynamically adjusted and optimized based on path environmental characteristics to evaluate path quality. A unified standard for scoring each path is integrated through a multi-objective fitness function, balancing multiple objectives such as path smoothness, path threat probability, and range efficiency. This achieves a balance among multiple objectives. Simultaneously, path smoothness is used to adaptively reduce path threat probability, enabling adaptive adjustment of the path itself. This coordinates and achieves safe, smooth, and efficient path planning, establishing an intelligent fitness adjustment mechanism capable of dynamically balancing multi-objective conflicts.

[0093] S40: Perform hierarchical particle swarm optimization iteration. In each iteration, the control parameters are adaptively adjusted based on the current particle swarm distribution characteristics, and the path node positions of the particles are updated.

[0094] In this embodiment, candidate flight paths are optimized in stages. The first step in optimization is to determine the distribution of candidate paths. Based on the distribution of candidate paths, path smoothness, path threat, and flight distance are flexibly adjusted, and the specific node positions of each candidate path are adjusted to optimize the candidate paths.

[0095] As shown in Figure 2, step S40 of the method provided in this embodiment includes:

[0096] In the first optimization phase, with global exploration as the goal, the range cost weight is adjusted first, and the distribution dispersion of the optimal path of an individual in the particle swarm is calculated. When the distribution dispersion is higher than a preset dispersion threshold, the adjustment range of the range cost weight is increased.

[0097] In the second optimization stage, based on the optimization results of the first optimization stage, with the goal of local refinement, the path smoothness weight and threat avoidance weight are adjusted first, the particle swarm aggregation degree is evaluated, and when the aggregation degree exceeds the preset aggregation threshold, the adjustment range of the path smoothness weight and threat avoidance weight is increased simultaneously.

[0098] In each iteration, the path node positions of the particles are updated based on the current control parameters.

[0099] Specifically, in the first optimization phase, with global exploration as the goal, the travel cost weight is adjusted first, and the distribution dispersion of the optimal paths of individuals in the particle swarm is calculated. When the distribution dispersion is higher than a preset dispersion threshold, the adjustment range of the travel cost weight is increased. A preset dispersion threshold is set based on historical data. If the dispersion is higher than the threshold, the adjustment range of the path travel cost weight is increased to encourage the exploration of shorter paths. Specifically, when calculating the distribution dispersion, firstly, the coordinates of all path nodes of the optimal path of an individual are extracted; secondly, the spatial center of the nodes is calculated: for the node coordinates of all optimal paths of an individual, the average value is calculated along the X, Y, and Z axes of the three-dimensional coordinate system to obtain the spatial center coordinates of the nodes; finally, the dispersion is calculated: the Euclidean distance from each node of the optimal path of an individual to the spatial center coordinates is calculated, and the standard deviation of all distances is taken, which is the distribution dispersion of the optimal path of the individual. The central tendency of the distribution dispersion is used to obtain the degree of clustering, and the degree of dispersion is used to obtain the degree of dispersion.

[0100] For example, if the distribution dispersion is 0.6, which is higher than the threshold of 0.5, the path distance cost weight needs to be increased to 0.5.

[0101] Furthermore, in the second optimization stage, based on the optimization results of the first optimization stage, with the goal of local refinement, the path smoothness weight and threat avoidance weight are adjusted first to evaluate the particle swarm aggregation degree. When the aggregation degree exceeds a preset aggregation threshold, the adjustment range of the path smoothness weight and threat avoidance weight is increased simultaneously. When calculating the particle swarm aggregation degree, firstly, based on the particle swarm updated in the first optimization stage, the spatial coordinates of all path nodes for each particle are extracted; secondly, the spatial center of the nodes is calculated; finally, the Euclidean distance from each path node of each particle to the corresponding spatial center of the node is calculated, and then the variance of all these distances is calculated, which is the particle swarm aggregation degree. The smaller the variance, the more concentrated the particles are in space, and the higher the aggregation degree.

[0102] For example, when the particle aggregation level reaches 0.7, the path smoothness weight and threat avoidance weight are increased from 0.5 and 0.7 respectively.

[0103] Furthermore, during each iteration, the particle's path node position is updated based on the current control parameters. Then, the control parameters are updated again based on the range cost weight obtained from global adjustment and the path smoothness weight and threat avoidance weight obtained from local adjustment, thereby adjusting the particle's path node position.

[0104] As shown in Figure 2, step S40 of the method provided in this embodiment further includes:

[0105] After each path node position is updated, a path feasibility repair mechanism is implemented to correct the projection of path nodes that exceed the flight corridor boundary or whose threat probability exceeds the limit.

[0106] An iterative convergence judgment mechanism is established, and the termination conditions of the optimization process are dynamically adjusted based on the improvement rate of the particle swarm fitness value and the stability of the particle distribution.

[0107] Specifically, after each path node position is updated, a path feasibility repair mechanism is implemented to correct the projection of path nodes that exceed the flight corridor boundary or whose threat probability exceeds the limit. Specifically, for nodes that exceed the flight corridor boundary, they are projected to the nearest boundary of the corridor; for nodes whose threat probability exceeds the limit, they are projected to the nearest safe location around the node, ensuring that all particles meet the spatial constraints.

[0108] For example, if a particle node exceeds the boundary of the flight corridor, the path feasibility repair mechanism detects that its threat probability is 1.0. It then calculates the nearest point to the convex hull surface and moves the node to that nearest point, reducing the threat probability to 0.

[0109] Furthermore, an iterative convergence judgment mechanism is established, dynamically adjusting the termination condition of the optimization process based on the improvement rate of the particle swarm fitness value and the stability of the particle distribution. The optimization iteration terminates when the termination condition is met, and the result is output.

[0110] Step S40 of the method provided in this application embodiment describes establishing an iterative convergence judgment mechanism, which dynamically adjusts the termination conditions of the optimization process based on the improvement rate of the particle swarm fitness value and the stability of the particle distribution, including:

[0111] Calculate the improvement rate of the global optimal path particle swarm fitness value within consecutive iteration cycles;

[0112] Calculate the particle swarm fitness variance and particle position similarity. When the particle swarm fitness variance is lower than a preset variance convergence threshold and the particle position similarity is higher than a preset similarity convergence threshold, the particle swarm distribution is determined to be stable.

[0113] Taking into account the current iteration count, the improvement rate of the particle swarm fitness value, and the stability of the particle distribution, the iteration terminates when any of the following conditions are met:

[0114] The maximum number of iterations has been reached.

[0115] The average improvement rate of particle swarm fitness values ​​over multiple consecutive iterations is lower than the adaptive convergence threshold.

[0116] The particle distribution stability continues to reach a stable state for more than the preset number of stable iterations.

[0117] Specifically, the improvement rate of the global optimal path particle swarm fitness value is calculated within consecutive iteration cycles. The global optimal path, i.e., the path with the highest fitness, is reached during the weighting of iterations n and n+1. The improvement rate is calculated as: [(Fitness of iteration (n+1) - Fitness of iteration n) / Fitness of iteration n] × 100%. The improvement rate reflects the particle update effect of the global optimal path particle swarm fitness value within consecutive iteration cycles; a higher value indicates a greater improvement effect of the path feasibility repair mechanism.

[0118] For example, after 5 iterations, the fitness values ​​of the globally optimal path are [1.02, 1.019, 1.0185, 1.0182, 1.0181]. Therefore, the improvement rate = [(1.019-1.02) / 1.02]≈0.0009, which means the improvement rate is 0.09%.

[0119] Furthermore, the particle swarm fitness variance and particle position similarity are calculated. When the particle swarm fitness variance is lower than a preset variance convergence threshold and the particle position similarity is higher than a preset similarity convergence threshold, the particle swarm distribution is considered to be stable. Specifically, when calculating the particle swarm fitness variance, first, the average fitness value of all particles in the current particle swarm is calculated; second, the square of the difference between each particle's fitness value and the average value is calculated; finally, the average of these squared values ​​is taken as the fitness variance. The smaller the variance value, the closer the particle swarm fitness values ​​are. When calculating the particle position similarity, the coordinate differences of corresponding nodes of all particles are compared according to the node order of each particle. The average of these corresponding node coordinate differences is calculated as the particle position similarity. The smaller the difference in particle position similarity, the higher the position similarity.

[0120] For example, the variance of the current fitness values ​​of all 100 particles is 0.0005, which is very low, indicating that the current fitness values ​​of the particles are similar. The structural similarity between all paths and the globally optimal path is calculated and then averaged, resulting in a particle position similarity of 0.92, which is very high, indicating a high degree of similarity in particle positions.

[0121] Furthermore, considering the current iteration count, the improvement rate of the particle swarm fitness value, and the stability of the particle distribution, the iteration terminates when any of the following conditions are met: the number of iterations reaches the maximum number of iterations; the average improvement rate of the particle swarm fitness value is lower than the adaptive convergence threshold for multiple consecutive iterations; or the particle distribution stability remains stable for more than the preset stable iteration count. Since these conditions are met, the algorithm determines that the iteration has converged, terminates the optimization, and outputs the results. This avoids unnecessary computation and efficiently ends the search process.

[0122] For example, the average improvement rate of 5 consecutive iterations is 0.01%, which is lower than the adaptive convergence threshold of 0.05%, satisfying condition 2. Therefore, it is determined that the iteration has converged and the optimization is terminated.

[0123] In this embodiment, when the particle swarm is scattered, a global adjustment is performed to locate the region where the global optimum may be found. Priority is given to adjusting the range cost weight and increasing its adjustment magnitude. Once the particle swarm gathers in a better region, the goal shifts to local refinement within this high-quality region. At this point, the path smoothness weight and threat avoidance weight are simultaneously increased. When updating particle positions, particles may exceed the flight corridor boundary or the threat probability may exceed the limit. The path feasibility repair mechanism returns the particles to the safe zone, projecting them to the nearest safe point or corridor boundary, ensuring that each iteration produces a physically feasible solution and avoiding invalid computation.

[0124] S50: When the convergence condition is met, output the globally optimal path as the final flight trajectory of the UAV and control the UAV to execute it.

[0125] In this embodiment of the application, when the convergence condition is met, the planning result is seamlessly connected to the UAV flight control system to form a closed loop, outputting the globally optimal path as the final flight trajectory of the UAV, and controlling the UAV to execute.

[0126] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0127] In this embodiment, firstly, by fusing multi-source terrain data such as digital elevation model, building distribution data, and no-fly zone coordinates, and using an exponential decay function to establish an obstacle probability field, the flight corridor boundary is generated. This unifies static obstacle avoidance and dynamic airspace management constraints into a spatial constraint model, ensuring the feasibility of the planned path and greatly improving the practicality and reliability of the method. Secondly, a sequence of path nodes is randomly generated within the flight corridor using Logistic chaotic mapping to ensure diversity. Then, based on the obstacle probability field, feasibility screening is performed to remove nodes whose threat probability exceeds a threshold. Finally, based on environmental dynamics, initial values ​​for the particle swarm control parameters are set. Chaotic mapping improves convergence speed and global search capability, effectively avoiding local optima, improving search efficiency, achieving comprehensive coverage, and ensuring diversity.

[0128] Secondly, a fixed-weighted linear weighting method is used to construct an objective function by combining path smoothness, corrected path threat probability, and path range efficiency. This function is dynamically adjusted and optimized based on path environment characteristics to evaluate path quality. A unified standard for scoring each path is integrated through a multi-objective fitness function, balancing multiple objectives such as path smoothness, path threat probability, and range efficiency. This achieves a balance among multiple objectives. Simultaneously, path smoothness is used to adaptively reduce path threat probability, enabling the path itself to self-adjust, coordinating to achieve safe, smooth, and efficient path planning, and establishing an intelligent fitness adjustment mechanism capable of dynamically balancing multi-objective conflicts. Furthermore, a fixed-weighted linear weighting method is used to construct an objective function by combining path smoothness, corrected path threat probability, and path range efficiency. This function is dynamically adjusted and optimized based on path environment characteristics to evaluate path quality. A unified standard for scoring each path is integrated through a multi-objective fitness function, balancing multiple objectives such as path smoothness, path threat probability, and range efficiency. To achieve a balance among multiple objectives, the path threat probability is adaptively reduced through path smoothness, enabling the path itself to self-adaptively adjust, thus coordinating safe, smooth, and efficient path planning, and establishing an intelligent adaptive adjustment mechanism that can dynamically balance conflicts among multiple objectives.

[0129] Compared to existing technologies, this application presents a UAV path planning method based on particle swarm optimization, which generates optimal flight paths in complex environments through adaptive parameter adjustment. The method emphasizes spatial constraint modeling, multi-objective optimization, and hierarchical iteration to improve path safety, smoothness, and efficiency. It implements a spatial constraint model for environmental threats, enhancing practicality and reliability. An adaptive adjustment mechanism based on path characteristics adaptively reduces the path threat probability, establishing an intelligent evaluation system capable of dynamically balancing multiple objectives. A hierarchical adaptive optimization strategy is employed for staged, adaptive parameter adjustments, leading to more efficient global optimal path finding. A path feasibility repair mechanism ensures the continuity and stability of the optimization process. An iterative convergence judgment mechanism intelligently determines the convergence state and dynamically adjusts, ensuring the stability of the globally optimal path output and providing an adaptive and executable optimal UAV path. This method achieves high-precision environmental modeling, multi-objective adaptive trade-offs, and intelligent search optimization for UAV path planning.

[0130] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0131] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An adaptive path planning method for unmanned aerial vehicles (UAVs) employing particle swarm optimization, characterized in that: The method includes: constructing a spatial constraint model for a UAV flight mission, wherein the spatial constraint model includes an obstacle distribution field and a flight corridor boundary; initializing a particle swarm based on the spatial constraint model, wherein each particle represents a candidate path, and the control parameters of the particle swarm include path smoothness weight, threat avoidance weight, and range cost weight; establishing a multi-objective fitness function by combining path smoothness, path threat probability, and range efficiency, and calculating the fitness value of each particle based on the particle swarm, including: analyzing the curvature change of the path node sequence, and calculating the average curvature change based on the rate of change of the angle between the vectors formed by three consecutive path nodes, as the path smoothness; extracting the maximum probability value of each path node in the obstacle probability field as the path threat probability, and adaptively reducing the path threat probability based on the path smoothness to obtain a corrected path threat probability, including: base A threat probability reduction coefficient is calculated based on the path smoothness, wherein the threat probability reduction coefficient and the path smoothness satisfy a negatively correlated exponential function relationship; the threat probability reduction coefficient is used to correct the path threat probability to obtain the corrected path threat probability; the path range efficiency is obtained by calculating the ratio of the total path length to the Euclidean distance from the mission start point to the mission end point; a multi-objective fitness function is constructed by combining the corrected path threat probability, path smoothness, and path range efficiency using a weighted summation method; the fitness value of each particle in the particle swarm is calculated based on the multi-objective fitness function; hierarchical particle swarm optimization iteration is performed, and in each iteration, the control parameters are adaptively adjusted based on the current particle swarm distribution characteristics, and the path node positions of the particles are updated; when the convergence condition is met, the globally optimal path is output as the final flight trajectory of the UAV, and the UAV is controlled to execute it.

2. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 1, characterized in that, Constructing a spatial constraint model for UAV flight missions includes: acquiring multi-source terrain data of the target mission area and constructing an obstacle probability field based on the multi-source terrain data, wherein the multi-source terrain data includes a digital elevation model, building distribution data, and no-fly zone coordinates; generating flight corridor boundaries using a fast exploratory random tree algorithm based on the mission start point, mission end point, and the obstacle probability field; and establishing a spatial constraint model that includes static obstacle constraints and dynamic airspace constraints by combining the obstacle probability field and the flight corridor boundaries.

3. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 2, characterized in that, Constructing an obstacle probability field based on the multi-source terrain data includes: performing terrain gradient analysis on the digital elevation model to identify areas with slopes greater than a preset slope threshold as terrain obstacle zones; performing three-dimensional convex hull modeling on the building distribution data to calculate the influence range of each building; establishing polygonal constraint boundaries based on the no-fly zone coordinates to determine the absolute no-fly zone; setting different safety distance thresholds based on the terrain obstacle zone, the influence range of buildings, and the absolute no-fly zone; establishing an obstacle threat probability distribution model using an exponential decay function, wherein the threat probability of the obstacle center area is 1, decreasing to 0 with increasing safety distance according to a preset decay coefficient; and fusing the threat probability distributions of the terrain obstacle zone, the influence range of buildings, and the absolute no-fly zone to generate a unified obstacle probability field.

4. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 1, characterized in that, Initializing the particle swarm based on the spatial constraint model includes: randomly generating an initial path node set within the flight corridor boundary using a Logistic chaotic mapping based on the coordinates of the mission start and end points; performing feasibility screening on the initial path node set based on the obstacle probability field, eliminating path nodes whose threat probability exceeds a preset threat probability threshold; setting initial values ​​for the control parameters of the particle swarm, wherein the path smoothness weight is determined based on the curvature variance of the initial path nodes, the threat avoidance weight is determined based on the maximum threat value of the obstacle probability field, and the flight cost weight is determined based on the Euclidean distance from the mission start to the mission end; and constructing a path node sequence from the mission start to the mission end by selecting the path nodes that meet the requirements, wherein each path node sequence corresponds to one particle, and multiple particles together constitute the particle swarm.

5. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 1, characterized in that, A hierarchical particle swarm optimization iteration is performed. In each iteration, the control parameters are adaptively adjusted based on the current particle swarm distribution characteristics, and the path node positions of the particles are updated. This includes: in the first optimization stage, with global exploration as the goal, the range cost weight is adjusted first, and the distribution dispersion of the optimal path of each individual in the particle swarm is calculated. When the distribution dispersion is higher than a preset dispersion threshold, the adjustment range of the range cost weight is increased. In the second optimization stage, based on the optimization results of the first optimization stage, with local refinement as the goal, the path smoothness weight and threat avoidance weight are adjusted first, and the particle swarm aggregation degree is evaluated. When the aggregation degree exceeds a preset aggregation threshold, the adjustment range of the path smoothness weight and threat avoidance weight is increased simultaneously. During each iteration, the path node positions of the particles are updated based on the current control parameters.

6. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 1, characterized in that, The process involves performing hierarchical particle swarm optimization iterations. In each iteration, the control parameters are adaptively adjusted based on the current particle swarm distribution characteristics, and the path node positions of the particles are updated. The process also includes: implementing a path feasibility repair mechanism after each path node position update to perform projection correction on path nodes that exceed the flight corridor boundary or have excessive threat probability; and establishing an iterative convergence judgment mechanism to dynamically adjust the termination conditions of the optimization process based on the improvement rate of the particle swarm fitness value and the stability of the particle distribution.

7. The adaptive path planning method for UAVs using particle swarm optimization as described in claim 6, characterized in that, An iterative convergence judgment mechanism is established, which dynamically adjusts the termination conditions of the optimization process based on the improvement rate of particle swarm fitness values ​​and particle distribution stability. This includes: calculating the improvement rate of the global optimal path particle swarm fitness values ​​within consecutive iteration cycles; calculating the particle swarm fitness variance and particle position similarity; determining that the particle swarm distribution tends to stabilize when the particle swarm fitness variance is lower than a preset variance convergence threshold and the particle position similarity is higher than a preset similarity convergence threshold; and terminating the iteration when any of the following conditions are met, considering the current iteration count, the improvement rate of particle swarm fitness values, and particle distribution stability: the number of iterations reaches the maximum number of iterations; the average improvement rate of particle swarm fitness values ​​is lower than the adaptive convergence threshold for multiple consecutive iteration cycles; or the particle distribution stability continuously reaches a stable state for more than a preset stable iteration count.

Images