A method and system for unmanned aerial vehicle path planning

Abstract

The application belongs to the technical field of unmanned aerial vehicle path planning, and particularly relates to an unmanned aerial vehicle path planning method and system, which comprises the following steps: acquiring three-dimensional space information of an environment where an unmanned aerial vehicle is located, and constructing an environment model containing obstacle distribution; setting energy consumption constraints, self-performance constraints and obstacle constraints in the flight process of the unmanned aerial vehicle, constructing a target function of path evaluation based on the constraints, generating candidate positions of particles by Monte Carlo random sampling on a probability density function, simultaneously taking an individual best position, a global best position and an average best position as a guide path when the particle positions are updated, introducing a chaotic sequence to generate a chaotic disturbance offset, and superimposing the chaotic disturbance offset to the particle position updating process, introducing a simulated annealing algorithm to perform local fine search on a current optimal solution in a late iteration stage, and stopping until a maximum iteration number is reached. The application can effectively improve the search capability of the unmanned aerial vehicle for a path in a three-dimensional space, and then safely and accurately reach a target point.

Description

Technical Field

[0001] This application belongs to the field of UAV path planning technology, specifically relating to a UAV path planning method and system. Background Technology

[0002] As a new force in the new era of technology, drones have high flexibility and rapid response capabilities, and have been widely used in disaster relief and mission execution. Many classic methods, such as graph search and heuristic optimization, have been applied to the drone path planning problem.

[0003] While current path planning methods have fast search capabilities and computational performance, they often suffer from problems such as getting trapped in local optima, slow convergence, premature convergence, and low algorithm diversity when dealing with complex real-world problems. Therefore, how to perform path planning for UAVs is of great significance in the field of UAV control optimization. Summary of the Invention

[0004] This application provides a method and system for UAV path planning, which solves the problems of existing technologies in dealing with UAV path planning problems, such as easy getting trapped in local optima, low convergence accuracy, poor search ability, low robustness and low algorithm complexity.

[0005] The first aspect of this application provides a method for unmanned aerial vehicle (UAV) path planning, including: Acquire three-dimensional spatial information of the environment in which the drone is located, and construct an environmental model that includes the distribution of obstacles; Energy consumption constraints, self-performance constraints, and obstacle constraints are set during the flight of the UAV. Based on the cumulative sum of the cost functions corresponding to each constraint and the predetermined weight coefficients, the objective function for path evaluation is constructed. An improved chaotic quantum particle swarm optimization algorithm is used for path search: each candidate path is encoded as a particle in the search space, and the state of each particle is described by a probability density function. Candidate positions of the particles are generated by Monte Carlo random sampling of the probability density function. When updating particle positions, the individual best position, the global best position, and the average best position are used as guiding paths, and the influence weight of each path is allocated by a balancing parameter. The fitness values ​​of all particles are calculated to determine the individual best position and the global best position of each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are superimposed on the particle position update process. It is determined whether the maximum number of iterations has been reached. If it has, the search result is output and the algorithm ends; otherwise, the particle state is updated iteratively, and a simulated annealing algorithm is introduced in the later stage of the iteration to perform a local fine search on the current optimal solution until the maximum number of iterations is reached, and the flight path that satisfies all constraints and optimizes the objective function value is obtained.

[0006] Furthermore, energy consumption constraints are set during the drone's flight, including: Based on the relationship between aerodynamic formulas and energy consumption, an energy consumption model under displacement is constructed. Based on the energy consumption model under displacement, construct the energy consumption model of the UAV under different flight states; The flight path is divided into multiple segments, and a maximum range is set. The energy consumption of each segment is expressed as a function of the corresponding time when the UAV is in each flight state within that segment. The energy consumption of each segment is accumulated to obtain the total energy consumption. Set a maximum energy consumption threshold for the drone. If the calculated total energy consumption exceeds the maximum energy consumption threshold, then apply a penalty to the particle's fitness value.

[0007] Furthermore, the self-performance constraints include velocity constraints, acceleration constraints, and flight attitude constraints; The speed constraint is a penalty when the drone's flight speed exceeds the maximum speed limit; The acceleration constraint is a penalty term when the UAV's acceleration exceeds the maximum tolerable acceleration or is less than the maximum deceleration; The flight attitude constraint is a penalty when the pitch and roll angles of the UAV exceed the corresponding maximum angle limits.

[0008] Furthermore, the obstacle constraint includes: The obstacle constraint is a penalty term when the minimum distance from the UAV to the obstacle is less than the safe distance threshold; wherein, the calculation of the minimum distance includes: the minimum value of the vertical distance from the current position of the UAV to the surface of each obstacle, and then subtracting the equivalent radius of the UAV; The calculation of the safe distance threshold includes: multiplying the maximum diameter of the drone by a first safety consideration coefficient to obtain the physical size safety margin; Multiply the minimum reliable detection distance by the second security consideration coefficient to obtain the perception security margin; Multiply the minimum turning radius of the drone by the third safety consideration factor to obtain the maneuver safety margin; The safety distance threshold is obtained by summing the physical size safety margin, the perception safety margin, the maneuver safety margin, and the accuracy error.

[0009] Furthermore, the cost function corresponding to each constraint is used to quantify the energy consumption constraint, self-performance constraint, and obstacle constraint into a comparable path evaluation value, normalize each constraint, and impose corresponding penalties on candidate paths that violate the constraints, so as to guide the particle swarm to search for low-cost regions that satisfy all constraints.

[0010] Furthermore, particle position updates also include: Generate a first random number that is uniformly distributed between 0 and 1. This first random number is used to perform a weighted summation of the current particle's individual best position and the population's global best position to obtain a weighted combined position. A second random number is generated that is uniformly distributed between 0 and 1. The distance weighting factor is obtained by taking the natural logarithm of the second random number and calculating its absolute value. The distance weighting factor is used to calculate the distance between the current position of the particle and the weighted combination position. A third random number is generated that is uniformly distributed between zero and one. When the third random number is less than 0.5, the candidate new position is centered on the weighted combination position, and a random offset whose amplitude is jointly controlled by the balance parameter and the distance weight factor is superimposed in the direction of the distance from the current position of the particle to the weighted combination position. When the third random number is greater than or equal to 0.5, the candidate new position is centered on the weighted combination position, and a random offset whose amplitude is jointly controlled by the balance parameter and the distance weight factor is superimposed in the direction of the distance from the average optimal position to the weighted combination position.

[0011] Furthermore, a chaotic sequence is introduced to generate chaotic perturbation offsets which are then superimposed on the particle position update process, including: The generation rule for the chaotic sequence is as follows: when the current value of the chaotic sequence is less than 0.5, the next value is twice the current value of the chaotic sequence; when the current value of the chaotic sequence is greater than or equal to 0.5, the next value is the difference between 2 multiplied by 1 and the current value of the chaotic sequence. The chaotic perturbation amplitude is calculated based on the ratio of the current iteration number to the maximum iteration number. The chaotic perturbation amplitude decreases linearly from a preset initial value with the number of iterations, and approaches zero at the end of the iteration. Multiply the current value of the chaotic sequence by the difference in the search boundary range, and then multiply by the chaotic perturbation amplitude of the current iteration to obtain the chaotic perturbation offset.

[0012] Furthermore, the process of introducing chaotic sequences to generate chaotic perturbation offsets and superimposing them on the particle position update process also includes: All particles in the population are divided into exploration and development groups according to their quantity ratio; The positions of the two groups of particles are updated independently along each coordinate axis in three-dimensional space. Each group of particles uses the current best position of the individual particle as a guiding reference, and generates candidate new positions by superimposing quantum search offset and chaotic perturbation offset. The quantum search offset is determined by multiplying the balance parameter by the distance between the current position of the particle and the average best position, and then by the distance weighting factor. The exploration group performs a global search by superimposing a chaotic perturbation offset of full magnitude, while the development group performs a local search by superimposing a chaotic perturbation offset of reduced magnitude.

[0013] Furthermore, in the later stages of iteration, a simulated annealing algorithm is introduced to perform a local refinement search for the current optimal solution, including: An elite strategy is adopted to select the particles with the best fitness from the current population as elite individuals after each iteration; A new candidate position is generated by superimposing a random perturbation term on the current position of the elite individual. The random perturbation term follows a standard normal distribution, and the perturbation amplitude is controlled by the current annealing temperature. The annealing temperature decreases linearly from a preset initial value to zero with the number of iterations. Calculate the fitness value of the candidate new position. When the fitness of the candidate new position is better than that of the original elite individual position, the candidate new position is accepted unconditionally and the elite individual is updated. When the fitness of the candidate new position is worse than that of the original elite individual position, the candidate new position is accepted with a preset probability. The preset probability is determined by a negative exponential function of the fitness difference divided by the current annealing temperature. The simulated annealing process is continuously iterated until the annealing temperature drops to the preset termination temperature. The elite individual that is finally retained is then output as the global optimal solution.

[0014] A second aspect of this application provides a drone path planning system, comprising: The environmental perception module is used to acquire three-dimensional spatial information of the environment in which the UAV is located and to construct an environmental model that includes the distribution of obstacles. The path constraint construction module sets the energy consumption constraints, self-performance constraints, and obstacle constraints during the UAV's flight. Based on the cumulative sum of the cost functions corresponding to each constraint and the predetermined weight coefficients, it constructs the objective function for path evaluation. The path search module encodes each candidate path as a particle in the search space, describing the state of each particle with a probability density function. Candidate positions for the particles are generated by Monte Carlo random sampling of the probability density function. During particle position updates, the individual best position, global best position, and average best position are used as guiding paths, and the influence weights of each path are allocated through a balancing parameter. The fitness values ​​of all particles are calculated to determine the individual best position and global best position for each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are then superimposed on the particle position update process. The module determines whether the maximum number of iterations has been reached. If so, the search result is output and the algorithm terminates; otherwise, the particle state continues to be iterated and updated. In the later stages of iteration, a simulated annealing algorithm is introduced to perform a local refinement search on the current optimal solution until the maximum number of iterations is reached, yielding a flight path that satisfies all constraints and optimizes the objective function value.

[0015] Compared with the prior art, the embodiments of this application have at least the following beneficial effects: This application leverages the powerful global search capability of the particle swarm optimization (PSO) algorithm by introducing chaotic sequences into it. This effectively explores the solution space. Furthermore, the introduction of nonlinear chaotic sequences avoids the simplistic nature of traditional PSO update factors, enhancing search capabilities in complex environments. It further prevents population stagnation and local optima, maintaining a balanced exploration process. This not only avoids premature convergence caused by local optima in swarm optimization algorithms but also improves the information processing capabilities of UAV swarms in complex environments, increasing algorithm diversity and accuracy during population evolution. This application applies the improved quantum particle swarm optimization algorithm to UAV flight path planning, effectively enhancing the UAV's path search capability in three-dimensional space, thereby enabling it to safely and accurately reach the target location. Attached Figure Description

[0016] Figure 1 A flowchart of a UAV path planning method provided in this application embodiment; Figure 2 A comparison diagram of the three-dimensional flight paths of the UAV path planning method of this application with the dung beetle and gray wolf composite algorithm and the fusion of bat and Creer group algorithm provided for the embodiments of this application; Figure 3 A comparison chart showing the optimal fitness values ​​of the UAV path planning method provided in this application, the dung beetle and gray wolf algorithm, and the fusion of bat and Creer group algorithm, provided for embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] See Figure 1 As shown, this application provides a method for unmanned aerial vehicle (UAV) path planning, including: S101: Obtain three-dimensional spatial information of the environment in which the drone is located, and construct an environmental model including the distribution of obstacles; S102, set the energy consumption constraints, self-performance constraints and obstacle constraints during the flight of the UAV, and construct the objective function of path evaluation based on the cumulative sum of the cost function corresponding to each constraint and the predetermined weight coefficient; S103, an improved chaotic quantum particle swarm optimization algorithm is used for path search: each candidate path is encoded as a particle in the search space, and the state of each particle is described by a probability density function. Candidate positions of particles are generated by Monte Carlo random sampling of the probability density function. When updating particle positions, the individual best position, the global best position, and the average best position are used as guiding paths, and the influence weight of each path is allocated by a balancing parameter. The fitness values ​​of all particles are calculated to determine the individual best position and the global best position of each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are superimposed on the particle position update process. It is determined whether the maximum number of iterations has been reached. If it has, the search result is output and the algorithm ends. Otherwise, the particle state is updated iteratively, and a simulated annealing algorithm is introduced in the later stage of the iteration to perform a local fine search on the current optimal solution until the maximum number of iterations is reached, and the flight path that satisfies all constraints and optimizes the objective function value is obtained.

[0019] In step S101, the UAV is equipped with a lidar system for environmental detection. The lidar acquires three-dimensional spatial information of the UAV's environment by emitting laser beams and receiving reflected signals. Specifically, the lidar system emits multiple laser pulses into the surrounding environment. When the laser pulses encounter obstacles, they are reflected, and the reflected signals are captured by the lidar's signal receiver. By calculating the time difference between the laser pulse emission and reception, and combining this with the speed of light constant, the distance information of the lidar to various points on the obstacle surface can be calculated, generating three-dimensional point cloud data of the environment.

[0020] The acquired environmental information is used to create a 3D model, specifically modeling the UAV flight environment as a 3D voxel map. The voxel map divides the 3D space into several small cube units (voxels) of equal size, and assigns an occupancy probability value to each voxel unit to characterize whether the voxel unit is occupied by an obstacle.

[0021] After completing the environmental modeling, a 3D obstacle model is constructed in the 3D voxel map based on the acquired environmental information to simulate the drone's flight path environment. The obstacle model includes the following types: Geometric obstacle models: These models approximate obstacles with regular shapes in the environment using regular geometric shapes, including but not limited to cuboids, spheres, cylinders, and cones. The parametric representation of geometric obstacles is as follows: for cuboid obstacles, it is determined by the coordinates of the center point and the dimensions of length, width, and height; for sphere obstacles, it is determined by the coordinates of the center point and the radius; for cylinder obstacles, it is determined by the coordinates of the center point, the radius of the base, and the height. Geometric obstacle models are suitable for simulating static obstacles with regular shapes, such as buildings, tree trunks, and utility poles.

[0022] Irregularly Shaped Obstacles: For obstacles with irregular shapes in the environment, a polyhedral mesh model or a triangular patch model is used for detailed modeling. The construction method for irregularly shaped obstacles is as follows: First, the point cloud of the obstacle surface acquired by LiDAR is triangulated to generate a surface mesh model composed of multiple triangular patches; then, the mesh model is simplified to reduce the number of patches while maintaining the main shape features of the obstacle, thereby reducing computational complexity. Irregularly shaped obstacles are suitable for simulating obstacles with irregular shapes such as rocks, landslides, and complex building outlines.

[0023] Dynamic Irregular Shape Obstacles: For obstacles with time-varying characteristics in the environment, dynamic models are used for modeling and real-time updates. Dynamic obstacles include moving drones, flocks of birds, and moving vehicles. The modeling method for dynamic irregular shape obstacles is as follows: within each time step, the position and shape changes of the dynamic obstacle are detected in real time using LiDAR; the motion state of the dynamic obstacle is predicted and estimated using tracking algorithms such as Kalman filtering or particle filtering; the position and shape parameters of the obstacle model are updated based on the prediction results to ensure that the environmental model is dynamically updated over time. The motion models of dynamic obstacles include uniform linear motion models, uniformly accelerated motion models, and random motion models, and the appropriate motion model is selected according to the motion characteristics of different types of dynamic obstacles.

[0024] In step S102, after completing the environmental modeling and obstacle model construction in step S101, this step designs the various constraints that the UAV flight path needs to meet, and constructs the cost function and objective function for evaluating the quality of the path based on these constraints.

[0025] Energy Constraints: During flight, the energy consumption of a drone primarily stems from the work done to overcome air resistance and the energy required to maintain flight under varying flight conditions. According to aerodynamic principles, the air resistance experienced by a drone during flight is proportional to air density, drag coefficient, frontal area, and the square of its flight speed. Energy consumption under displacement is represented as the cumulative work done by air resistance along the displacement path.

[0026] Unmanned aerial vehicles (UAVs) exhibit different motion states during different flight phases, primarily including horizontal cruise, vertical climb, vertical descent, and turning maneuvers. Energy consumption in each state is determined by the corresponding dynamic model.

[0027] Assuming the drone's flight path is divided into N segments, each segment contains a combination of different flight states, and the drone has a preset maximum range, the energy consumption of each segment is expressed as a function of the corresponding time of the drone in each flight state within that segment. By summing the energy consumption of each segment, the total energy consumption of the entire flight path can be obtained.

[0028] A maximum energy consumption threshold is set for the drone. If the total energy consumption of a candidate path exceeds this threshold, it indicates that the path does not meet the energy consumption constraint requirements. To effectively eliminate paths that do not meet the energy consumption constraints during the path search process, a penalty is imposed on particles that violate the energy consumption constraints. Specifically, when calculating the particle fitness value, if the total energy consumption of the path represented by the particle exceeds the maximum energy consumption threshold, a penalty term is added to the original fitness value. The penalty term is determined by multiplying a preset penalty coefficient greater than 1 by the excess amount between the total energy consumption and the maximum energy consumption threshold.

[0029] Based on aerodynamic formulas and the relationship between energy consumption, the energy consumption model under displacement with respect to air resistance and energy consumption can be calculated as follows: , In the formula, displacement Energy consumption under, air density, The drag coefficient, For windward area, For drone speed, Let be the flight time of the drone; considering different flight states of the drone, the energy consumption models are calculated as follows: , , , , In the formula, , , , These represent energy consumption under four different conditions. This refers to the energy consumed by the drone while it is hovering. This refers to the energy consumed by the drone during horizontal cruise. This refers to the energy consumed by the drone during vertical climb. This refers to the energy consumed by the drone during vertical descent. For the quality of drones, It is the acceleration due to gravity. The drag coefficient is the value of the vehicle in a hovering state. This refers to the speed of the drone while it is hovering. This refers to the flight time of the drone while hovering. The drag coefficient is the value under horizontal cruising conditions. The speed of the drone during horizontal cruise. This refers to the flight time of the drone in horizontal cruise mode. The drag coefficient during vertical ascent. The speed of the drone during vertical climb. This refers to the acceleration of the drone during vertical climb. This refers to the flight time of the drone during vertical climb. The drag coefficient during vertical descent. The speed of the drone during vertical descent. Let be the flight time of the drone in a vertical descent state; assuming the drone's flight path is divided into... The maximum range is Then the first The energy consumption of a segment is expressed as follows: .

[0030] In the formula Indicates the first Total energy consumption of the segment Indicates the first The corresponding time of the UAV in each state described in the paragraph; thus, the total energy consumption expression is derived as follows: , Based on the total energy consumption, design energy consumption constraints.

[0031] The maximum energy consumption of the drone is determined to be If the calculated total energy consumption Therefore, the fitness value of the particle needs to be penalized, and the energy consumption constraint is as follows: , In the formula, This is the penalty coefficient for energy consumption constraints. The function value is the energy consumption constraint, if the original fitness value is... The fitness value after the penalty is The expression is as follows: .

[0032] The inherent performance constraints of drones include: speed constraints, acceleration constraints, and flight attitude constraints; The speed constraint is a penalty when the drone's flight speed exceeds the maximum speed limit; The acceleration constraint is a penalty term when the UAV's acceleration exceeds the maximum tolerable acceleration or is less than the maximum deceleration; The flight attitude constraints are penalties for when the UAV's pitch and roll angles exceed the corresponding maximum angle limits. These constraints ensure that the planned flight path is within the UAV's physical capabilities.

[0033] Among these constraints, speed is a limiting factor. The UAV's flight speed is restricted by the maximum output capacity of its propulsion system and the safety limitations of its aerodynamic structure, and cannot exceed a preset maximum flight speed. For each segment of the candidate path, the average or instantaneous speed of the UAV on that segment is calculated. If the speed value exceeds the maximum flight speed, a corresponding penalty is applied. The penalty for the speed constraint is the amount of speed exceeding the limit multiplied by the corresponding penalty coefficient.

[0034] Acceleration constraints: The acceleration of a UAV is limited by the thrust-to-weight ratio of its propulsion system and structural strength, and cannot exceed a preset maximum acceleration. Simultaneously, the deceleration of the UAV cannot be less than a preset maximum deceleration (i.e., the limit of braking capacity). The penalty for acceleration constraints is the amount by which acceleration or deceleration exceeds its corresponding limit multiplied by the corresponding penalty coefficient. Acceleration constraints may also include jerk (rate of change of acceleration) constraints to ensure smooth flight trajectory and reduce drastic changes in UAV attitude.

[0035] Flight attitude constraints: The flight attitude of a UAV includes pitch and roll angles. Pitch angle is the angle between the UAV's longitudinal axis and the horizontal plane, and roll angle is the angle between the UAV's lateral axis and the horizontal plane. The UAV's pitch and roll angles are limited by the attitude control capabilities and aerodynamic stability boundaries of its flight control system and cannot exceed their respective preset maximum angle limits. The penalty for flight attitude constraints is the amount by which the pitch or roll angle exceeds its corresponding maximum angle limit multiplied by the corresponding penalty coefficient.

[0036] Obtain the inherent performance constraints of the drone. Represented as: , In the formula, , and These represent the velocity constraint function, acceleration constraint function, and flight attitude constraint function, respectively; their corresponding expressions are shown below: , , , in, The value of the velocity constraint function. It is the penalty coefficient for speed constraints. It is an exponential factor for the speed limit penalty. It is an exponential factor for the speed lower limit penalty, typically with a value of 2 or higher. For drones in the The actual flight speed of each segment This is the maximum speed limit value. This is the minimum speed limit value; For the acceleration constraint function value, For acceleration, The penalty coefficient for acceleration constraints, and These are the maximum acceleration and maximum deceleration that the drone can withstand. An exponential factor for the penalty on the upper limit of acceleration; The function value of the pitch angle constraint. is the penalty coefficient for attitude constraints. The pitch angle, For roll angle, The maximum pitch angle, Minimum pitch angle, For the maximum roll angle, Minimum roll angle, The function value of the roll angle constraint. The exponential factor for the pitch angle constraint penalty. This is an exponential factor for the roll angle constraint penalty.

[0037] Obstacle Constraints: To ensure the flight safety of the drone, the planned flight path must maintain a sufficient safe distance from all obstacles in the environment. Obstacle constraints require that the minimum distance between the drone and any obstacle during flight is not less than a preset safe distance threshold.

[0038] The calculation method for the minimum distance from the drone to an obstacle depends on the type of obstacle. For geometrically shaped obstacles, the minimum distance is the shortest Euclidean distance between the drone's surface and the obstacle's surface. Taking a spherical obstacle as an example, the minimum distance is the distance from the drone's center to the sphere's center minus the sphere's radius and then minus the drone's equivalent radius. For polyhedral obstacles, the minimum distance is the minimum distance from the drone's center to all triangular faces minus the drone's equivalent radius. For dynamic obstacles, the calculation of the minimum distance needs to consider the obstacle's predicted trajectory over a future period, taking the minimum distance between the drone and the obstacle within the predicted timeframe. In this embodiment, the calculation method is as follows: obtain the three-dimensional coordinates and unit normal vector of each surface of the obstacle, calculate the vertical distance from the drone's current position to each obstacle surface, take the minimum of all vertical distances, and then subtract the drone's equivalent radius to obtain the minimum distance from the drone to the obstacle. The safe distance threshold is determined by considering the physical dimensions of the drone, the reliable detection range of the sensors, the tracking accuracy of the control system, and the path tracking error. Specifically, the safe distance threshold is determined by the following parameters: the maximum diameter of the drone, the minimum reliable detection range of the drone's sensors, the accuracy error, and the minimum turning radius of the drone. These parameters are weighted and combined according to preset safety consideration coefficients.

[0039] If the minimum distance from the drone to an obstacle on a candidate path is less than the safe distance threshold, it indicates that the path does not meet the safety obstacle avoidance requirements. A corresponding penalty will be imposed on the path, and the penalty will increase as the distance shortfall increases.

[0040] Design obstacle constraints: , In the formula, The function value for the obstacle constraint. The penalty coefficient for obstacle constraints. This represents the minimum distance from the UAV to the obstacle in the path planning algorithm. The safe distance thresholds in drone path planning are expressed as follows: , , in, Let be the normal vector of the obstacle surface. , , All are safety considerations. The maximum diameter of the drone For accuracy error, The current coordinates of the drone. The three-dimensional coordinates of the obstacle surface. This represents the minimum turning radius of the UAV in the path planning algorithm. This represents the minimum reliable detection distance for drone sensors.

[0041] Specifically, the safe distance threshold is obtained by multiplying the maximum diameter of the UAV by a first safety consideration coefficient to obtain a physical size safety margin; by summing the measurement error, positioning error, and tracking error of the sensor and the control system to obtain an accuracy error; by multiplying the minimum reliable detection distance by a second safety consideration coefficient to obtain a perception safety margin; by multiplying the minimum turning radius of the UAV by a third safety consideration coefficient to obtain a maneuver safety margin; and by summing the physical size safety margin, the perception safety margin, the maneuver safety margin, and the accuracy error to obtain the safe distance threshold. In one embodiment, the total cost function is a weighted sum of the cost functions corresponding to each constraint. This sum is used to quantify the energy consumption constraint, self-performance constraint, and obstacle constraint into comparable path evaluation values, normalize each constraint, and impose corresponding penalties on candidate paths that violate the constraints, thereby guiding the particle swarm to search for low-cost regions that satisfy all constraints. Total Cost Function Represented as: , In the formula, , and This represents the proportion of each constraint in the total cost function, and , This is the cost function corresponding to the energy consumption constraint. The cost function corresponding to its own performance constraints. This is the cost function corresponding to the obstacle constraint.

[0042] The cost function quantifies the degree of constraint violation on a path into a comparable penalty value and incorporates it into the particle fitness evaluation system. In each iteration, particles that violate constraints suffer additional penalties and thus have higher fitness values ​​(worse values ​​in minimization problems), leading to their elimination in the competition between individual best and swarm best. Particles that satisfy all constraints gain a competitive advantage due to the lack of penalty and gradually dominate the population during iterations. As the number of iterations increases, the entire particle swarm migrates from the infeasible region to the feasible region under the influence of the penalty mechanism, eventually converging on the path that satisfies all constraints and has the optimal objective function value. Therefore, the penalty mechanism of the cost function continuously guides the search direction of the particle swarm, enabling the algorithm to automatically search for low-cost regions that satisfy all constraints without explicitly handling the constraints.

[0043] In one embodiment, for the cost function corresponding to each constraint analyzed in the UAV path planning method, predetermined weight coefficients are obtained. Based on the cumulative sum of the cost function corresponding to each constraint and the corresponding predetermined weight coefficient, the objective function of the UAV path planning method in the flight path is obtained, as follows: , In the formula , and All are predetermined weighting coefficients. This represents the objective function value.

[0044] In step S103, in the standard particle swarm optimization algorithm, the state of each particle is described by a deterministic position vector and velocity vector, and the particle's motion follows the velocity-displacement update equation within the framework of Newtonian mechanics. This embodiment employs a quantum particle swarm optimization algorithm, introducing the superposition principle and interference effects of quantum physics to re-describe the particle state. Under the quantum particle swarm framework, the state of each particle is no longer described by deterministic position and velocity vectors, but rather represented by a probability density function in the form of a wave function. Specifically, the position of each particle in three-dimensional space is described by a probability density function, which represents the probability distribution of the particle's occurrence at various positions in three-dimensional space. The specific position of the particle is not uniquely determined by a deterministic equation of motion, but is generated through random sampling under the constraint of the probability density function.

[0045] The central position of the probability density function is determined by the individual optimal position of the particle, the global optimal position of the population, and the average optimal position of the population. This design simulates the probability distribution characteristics of a particle under the superposition of multiple potential wells in quantum mechanics, enabling the particle to perform probabilistic searches on a global scale.

[0046] During population initialization, the population is first generated from... An initial population of particles. The initial position of each particle is randomly generated in a uniform distribution within the three-dimensional search space. The initial coordinates of each particle along each coordinate axis in three-dimensional space are determined, with each coordinate value randomly selected within the corresponding search boundary. The search boundary is determined by the size of the environment map.

[0047] In the quantum particle swarm optimization algorithm, in addition to the individual best position and the global best position, the concept of average best position is introduced. The average best position is defined as the arithmetic mean of the best positions of all individual particles in the current iteration along the coordinate axes in three-dimensional space.

[0048] The quantum superposition principle is introduced into this algorithm as follows: the candidate new position of each particle is no longer generated by a single deterministic path, but is jointly determined by multiple guiding benchmarks (individual best position, global best position, and average best position) through the superposition of probability amplitudes. In each iteration, candidate positions of particles are generated by Monte Carlo random sampling of the probability density function, thereby simulating the physical process of the quantum state collapsing from a superposition state to a deterministic eigenstate upon measurement.

[0049] Monte Carlo random sampling refers to generating a first random number that is uniformly distributed between 0 and 1 during the position update process of each particle. This first random number is then used to perform a weighted summation of the current particle's individual best position and the population's global best position to obtain a weighted combined position.

[0050] Represented as: , when When the value is large, the weighted combination position of the particles is more biased towards the individual optimal position, and the particles tend to rely on their own experience to search; when When the value is small, the weighted combination position is more biased towards the global optimal position, and the particles tend to rely on the collective experience to search. The randomness of the process makes each particle depend on individual and group experience to varying degrees at different iterations, thus enhancing the diversity of the population.

[0051] The parameters in the formula are defined as follows: For the first The particle in the first In the nth iteration The weighted combination position of the dimension The first random number, For the first The particle in the first In the nth iteration The weighted combination position of the dimension For the first In the nth iteration, the entire population is at the ... The global optimal position of dimensionality. This indicates the range of values.

[0052] Generate a second random number uniformly distributed between 0 and 1. Obtain the distance weighting factor by taking the natural logarithm and absolute value of this second random number. The value of this distance weighting factor is within... The distance is asymmetrically distributed within the range: when the second random number is close to 1, the distance weight factor is close to 0, and the particle performs a fine search near the guiding benchmark; when the second random number is close to 0, the distance weight factor tends to infinity, and the particle deviates significantly from the guiding benchmark to explore a large area. This distribution characteristic enables the algorithm to simultaneously perform local exploration and global exploration in a single iteration.

[0053] Generate a third random number that is uniformly distributed between 0 and 1. Based on the value of the third random number, select the distance between the current position of the particle and the weighted combination position as the offset reference, or the distance between the average optimal position and the weighted combination position as the offset reference.

[0054] When the third random number is less than 0.5, the candidate new position of the particle is generated by superimposing a random offset in the direction of distance from the current position of the particle to the weighted combination position, with the weighted combination position as the center. When the third random number is greater than or equal to 0.5, the candidate new position is generated by superimposing a random offset whose amplitude is jointly controlled by the balance parameter and the distance weight factor in the direction of the distance from the average optimal position to the weighted combined position, with the weighted combined position as the center.

[0055] The two generation methods are selected by taking the value of a third random number, and each method has a 50% probability of being selected, simulating the behavior characteristics of the superposition of probability amplitudes of different paths in quantum interference effect.

[0056] The formula is expressed as follows: , , , , in, This refers to the third random number. No. The particle in the first In the 1st iteration Victor's new candidate position For the first The particle in the first In the nth iteration Victor's new candidate position For the first The particle in the first In the nth iteration The weighted combination position of the dimension The second random number, For balance parameters; The total number of particles in the population. The preset maximum value of the balance parameter, This represents the current iteration number. This is the preset minimum value for the balance parameter. i is the preset maximum number of iterations.

[0057] The aforementioned balancing parameters control the particle search step size, balancing the algorithm's global exploration capability and local exploitation capability. In the early stages of algorithm iteration, Larger values ​​result in larger particle search step sizes, enabling a wide-ranging global exploration across the entire search space and preventing the algorithm from prematurely getting trapped in local optima. In the later stages of algorithm iteration, Smaller values ​​result in smaller particle search steps, enabling finer local searches near the current optimal solution and improving convergence accuracy.

[0058] The balance parameter is adjusted by using an exponential function with the natural constant as the base to control the decay rate, so that the balance parameter gradually decays from the preset maximum value to the preset minimum value.

[0059] In each iteration, the fitness value calculation unit substitutes the best position of each particle in the population in the current iteration into the objective function constructed in step S102 to calculate the fitness value of each particle.

[0060] For paths that satisfy the constraints (i.e., all constraint violations are zero), the objective function value is calculated directly. The objective function value is calculated based on the constructed objective function, comprehensively considering multiple optimization objectives such as path length, flight time, energy consumption, and smoothness.

[0061] For paths that do not meet the constraints, after calculating the objective function value, it is also necessary to calculate additional constraint violation penalty values ​​based on the constructed cost function.

[0062] The final fitness value of a particle is obtained by summing the objective function value and the cost function value (i.e., the constraint violation penalty value). The fitness value is the sole criterion for evaluating the quality of a particle. In the minimization problem, a particle with a smaller fitness value represents a better path. In each iteration, the current fitness value of each particle is compared with its historical best fitness value. If it is smaller, the particle's individual best position is updated, and the best fitness value is updated to the current fitness value. This comparison mechanism ensures that each particle always retains the position with the best fitness value on its search trajectory.

[0063] After updating the individual best positions of all particles, the individual best positions of all particles are traversed and compared, and the position of the particle with the best (i.e., the smallest) fitness value is selected as the global best position for the current iteration.

[0064] The obtained fitness values ​​of each particle are used to first determine the global best position of the population in the current iteration. The specific method for determining the global best position is as follows: sort the individual best positions of all particles in ascending order of their fitness values; the position of the particle with the smallest fitness value is the global best position for the current iteration. If multiple particles have the same minimum fitness value, then the position of one of these particles is randomly selected as the global best position.

[0065] After determining the global optimum, the iteration record of the global optimum is updated. The global optimum gradually approaches the global optimal solution during the algorithm's search process, and its iteration trajectory reflects the algorithm's convergence progress.

[0066] This embodiment introduces a chaotic sequence during the population update process to generate chaotic perturbation terms and guide the generation of the average optimal position. The generation rule for chaotic sequences is as follows: when the current value of the chaotic sequence is less than 0.5, the next value is twice the current value of the chaotic sequence; when the current value of the chaotic sequence is greater than or equal to 0.5, the next value is the difference between 2 multiplied by 1 and the current value of the chaotic sequence. The chaotic perturbation amplitude is calculated based on the ratio of the current iteration number to the maximum iteration number. The chaotic perturbation amplitude decreases linearly from a preset initial value with the number of iterations, and approaches zero at the end of the iteration. Multiply the current value of the chaotic sequence by the difference in the search boundary range, and then multiply by the chaotic perturbation amplitude of the current iteration to obtain the chaotic perturbation offset.

[0067] The formula is expressed as follows: , The magnitude of the chaotic disturbance is calculated based on the ratio of the current iteration number to the maximum iteration number. , The generated chaotic perturbation term: , Exploration team location update: , Development team location update: , in, For the first The current value of the chaotic sequence generated in the next iteration. For the first The current value of the chaotic sequence generated in the next iteration. This represents the magnitude of the chaotic perturbation in the current iteration, used to control the size of the chaotic perturbation offset. For the first The particle in the first Chaotic perturbation term on dimension; For the first The particle in the first Updated on the wei platform Axial coordinates, For the first The particle in the first Updated on the wei platform Axial coordinates, For the first The particle in the first Updated on the wei platform Axial coordinates, The coordinates of the optimal position of the particle in three-dimensional space Axial coordinates, The optimal position for an individual particle in three-dimensional space. Axial coordinates, The optimal position for an individual particle in three-dimensional space. Axial coordinates, The average optimal position of the population along each coordinate axis in three-dimensional space. Axial coordinates, The average optimal position of the population along each coordinate axis in three-dimensional space. Axial coordinates, The average optimal position of the population along each coordinate axis in three-dimensional space. Axial coordinates, The second random number, for The difference in the search boundary range along the axis. for The difference in the search boundary range along the axis. for Difference in the search boundary range along the axis.

[0068] The chaotic perturbation amplitude is a control parameter that dynamically changes with the number of iterations. It is used to adjust the magnitude of the chaotic perturbation offset. The chaotic perturbation amplitude is calculated based on the ratio of the current iteration number to the maximum iteration number. The chaotic perturbation amplitude decreases linearly from a preset initial value with the number of iterations, and approaches zero at the end of the iteration.

[0069] Multiply the current value of the chaotic sequence by the difference in the search boundary range and then by the chaotic perturbation amplitude of the current iteration to obtain the chaotic perturbation offset. The chaotic perturbation offset is calculated independently in each coordinate axis direction of the three-dimensional space.

[0070] Axial direction: The chaotic perturbation offset along the axis is equal to the current chaotic sequence value multiplied by The difference in the search boundary range along the axis is then multiplied by the current chaotic perturbation amplitude, corresponding to the formula above: , .

[0071] Axial direction: The chaotic perturbation offset along the axis is equal to the current chaotic sequence value multiplied by The difference in the search boundary range along the axis is then multiplied by the current chaotic perturbation amplitude; corresponding to the formula above: ; Axial direction: The chaotic perturbation offset along the axis is equal to the current chaotic sequence value multiplied by The difference in the search boundary range along the axis is then multiplied by the current chaotic perturbation amplitude, corresponding to the formula above: .

[0072] After generating the chaotic perturbation offsets in each coordinate axis direction, they are superimposed on the particle position update process.

[0073] All particles in the population are divided into an exploration group and a development group according to their quantity ratio. In one example, 70% of the particles in the population are composed of the exploration group, which uses greater perturbation to perform global search and update iteration, while 30% of the particles are composed of the development group, which uses less perturbation to perform local search and update iteration.

[0074] The positions of the two groups of particles are updated independently along each coordinate axis in three-dimensional space. Each group of particles uses its current best position as a guiding reference, superimposing a quantum search offset and a chaotic perturbation offset to generate candidate new positions. The quantum search offset is determined by multiplying the balance parameter by the distance between the particle's current position and the average best position, and then by a distance weighting factor. Corresponding to the formula above, in... Taking the axial direction as an example: .

[0075] For the exploration group particles (which make up 70% of the total population), the particles are in The candidate new position along the axis is equal to the quantum search offset in that direction plus... The offset of the complete chaotic perturbation along the axis. The particle in... The candidate new position along the axis is equal to the quantum search offset in that direction plus... The offset of the complete chaotic perturbation along the axis. The particle in... The candidate new position along the axis is equal to the quantum search offset in that direction plus... The offset of the complete chaotic perturbation in the axial direction.

[0076] For the development group particles (representing 30% of the total population), the particles in The candidate new position along the axis is equal to the quantum search offset in that direction plus... The reduced chaotic perturbation offset along the axis is reduced by 0.5 times the full magnitude. The particle in... Axial direction and The same applies to the axial direction.

[0077] The chaotic perturbation offset is recalculated in each iteration and changes dynamically as the iteration progresses.

[0078] In one embodiment, a simulated annealing algorithm is introduced in the later stages of the iteration to perform a local fine search for the current optimal solution, including: An elite strategy is adopted to select the particles with the best fitness from the current population as elite individuals after each iteration; A new candidate position is generated by superimposing a random perturbation term on the current position of the elite individual. The random perturbation term follows a standard normal distribution, and the perturbation amplitude is controlled by the current annealing temperature. The annealing temperature decreases linearly from a preset initial value to zero with the number of iterations. Calculate the fitness value of the candidate new position. If the fitness of the candidate new position is better than that of the original elite individual position, the candidate new position is accepted unconditionally and the elite individual is updated. If the fitness of the candidate new position is worse than that of the original elite individual position, the candidate new position is accepted with a preset probability. The preset probability is determined by a negative exponential function of the fitness difference divided by the current annealing temperature. The simulated annealing process is continuously iterated until the annealing temperature drops to the preset termination temperature. The elite individual that is finally retained is then output as the global optimal solution.

[0079] In the iterative process of the particle swarm optimization algorithm, the population explores a large area globally in the early and middle stages of the iteration through quantum search and chaotic perturbation, gradually approaching the global optimum. However, in the later stages of the iteration, when the population has converged to the vicinity of the optimum, relying solely on the position update mechanism of the particle swarm optimization is insufficient for sufficiently fine-grained searches in local regions, resulting in limited accuracy of the final solution.

[0080] Simulated annealing (SAM) possesses the ability to perform fine-grained searches in local regions. By gradually decreasing the temperature parameter, it transitions the search process from coarse-grained to fine-grained, enabling in-depth local exploration near the current optimal solution. This embodiment introduces SAM in the later stages of iteration to perform a fine-grained local search on the current optimal solution. The local search capability of SAM compensates for the shortcomings of particle swarm optimization (PSO) in fine-grained convergence in the later stages, thereby improving the accuracy of the final planned path.

[0081] The annealing temperature gradually decreases with increasing iterations, and its update method is a linear decrease. In each simulated annealing iteration, the current annealing temperature... Calculate using the following formula: , Where 100 represents the initial temperature, This indicates the annealing temperature.

[0082] The timing of initiating the simulated annealing local search is determined by a preset late-stage triggering condition. In this embodiment, the late-stage triggering condition is when the number of iterations reaches 80% of the preset maximum number of iterations. That is, in the first 80% stage before the algorithm's iteration count reaches the maximum number of iterations, the algorithm only executes the main iteration process of the quantum particle swarm optimization, including quantum position updates, fitness calculations, chaotic perturbations, and population updates. When the number of iterations reaches 80% of the maximum number of iterations, the algorithm enters the late-stage iteration phase, at which point the simulated annealing local search process is initiated.

[0083] The reason for setting the trigger condition to 80% of the maximum number of iterations is that in the first 80% of the iterations, the population has gradually converged to the vicinity of the global optimum through a large number of global searches. At this point, the elite individuals in the population already have high quality and meet the initial conditions for fine-grained local searches. Starting simulated annealing too early may lead to local searches before the global optimum is determined, wasting computational resources. Starting it too late may result in insufficient iterations to complete fine-grained convergence.

[0084] Before the simulated annealing local fine search begins, an elite strategy is used to preserve the optimal positions of the particles with the best fitness in the current iteration. Specifically, after each iteration, a certain proportion of the particles with the best fitness are selected from the current population, and their optimal positions are separated from the main population and stored separately in an elite archive.

[0085] Elite individuals are selected based on their fitness value. Particles with lower fitness values ​​represent better paths and are given priority for selection as elite individuals. The elite retention rate is typically set to 5% to 10% of the total population. Too high a retention rate will result in too many elite individuals, increasing the computational burden of simulated annealing. Too low a retention rate may cause the best individuals to be lost during iteration.

[0086] Elite individuals do not participate in position update operations in subsequent regular iterations, thus preventing their superior genes from being destroyed during the iteration process. When simulated annealing is triggered in the later stages of iteration, the elite individual with the best fitness in the elite file is used as the initial solution for simulated annealing. This design ensures that the starting point of the simulated annealing local search is the optimal solution found in the entire search process, thereby improving the efficiency of the local search and the quality of the final solution.

[0087] In the simulated annealing local search process, the current position of the elite individual is used as the initial solution, and a new candidate position is generated by superimposing a random perturbation term on this position. The random perturbation term follows a standard normal distribution with a mathematical expectation of zero and a variance of 1. That is, the perturbation value is symmetrically distributed near zero, with a high probability of small perturbations and a low probability of large perturbations.

[0088] The three coordinate components of the new location are generated independently. The new position coordinates along the axis are equal to the elite individual's position in the x-axis direction. The current position coordinates along the axis are added to the annealing temperature and multiplied by a random number that follows a standard normal distribution. The new position coordinates along the axis are equal to the elite individual's position in the x-axis direction. The current position coordinates along the axis are added to the annealing temperature and multiplied by a random number that follows a standard normal distribution. The new position coordinates along the axis are equal to the elite individual's position in the x-axis direction. The current position coordinates along the axis are added to the annealing temperature and multiplied by a random number that follows a standard normal distribution.

[0089] The formula is expressed as: , , , in, express New position coordinates in the axial direction, Indicates the first An elite individual Current position coordinates along the axis. A row representing a standard normal distribution Column random number matrix, Indicates the number of dimensions in the search space; express New position coordinates in the axial direction, Indicates the first An elite individual Current position coordinates along the axis; express New position coordinates in the axial direction, Indicates the first An elite individual Current position coordinates along the axis.

[0090] The simulated annealing local search process continues after startup. In each iteration, the following operations are performed: calculating the current annealing temperature, generating new candidate positions, calculating the fitness value of the new positions, determining whether to accept the new positions, and updating the positions of elite individuals. The particle with the lowest fitness at the globally optimal position is selected, and the process is continuously updated iteratively to complete the selection of the optimal flight path.

[0091] Another aspect of this application provides a UAV path planning system that implements the UAV path planning method of the above embodiments, including: The environmental perception module is used to acquire three-dimensional spatial information of the environment in which the UAV is located and to construct an environmental model that includes the distribution of obstacles. The path constraint construction module sets the energy consumption constraints, self-performance constraints, and obstacle constraints during the UAV's flight. Based on the cumulative sum of the cost functions corresponding to each constraint and the predetermined weight coefficients, it constructs the objective function for path evaluation. The path search module employs an improved chaotic quantum particle swarm optimization algorithm for path search: each candidate path is encoded as a particle in the search space, and the state of each particle is described by a probability density function. Candidate positions of the particles are generated by Monte Carlo random sampling of the probability density function. When updating particle positions, the individual best position, the global best position, and the average best position are used as guiding paths, and the influence weight of each path is allocated by a balancing parameter. The fitness values ​​of all particles are calculated to determine the individual best position and the global best position of each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are superimposed on the particle position update process. It is determined whether the maximum number of iterations has been reached. If it has, the search result is output and the algorithm ends; otherwise, the particle state is updated iteratively, and a simulated annealing algorithm is introduced in the later stages of iteration to perform a local refinement search on the current optimal solution until the maximum number of iterations is reached, resulting in a flight path that satisfies all constraints and optimizes the objective function value.

[0092] The three-dimensional flight paths obtained using the above method or system are compared with those obtained using the dung beetle and gray wolf coincidence algorithm and the fusion of bat and Creer swarm algorithm. The comparison results are shown in [link to comparison]. Figure 2 As shown, the three-dimensional spatial information of the UAV's environment is consistent from start to finish. The obstacle distribution environment model is the same, which is equivalent to the dung beetle and gray wolf composite algorithm and the fusion of bat and Creer swarm algorithm. Due to the UAV's own dynamic constraints (such as minimum turning radius limit), the smoothness of the planned path directly affects the tracking accuracy of the flight control system. The generated smooth path reduces unnecessary heading changes and reduces the tracking burden of the flight control system. By reasonably balancing path length and obstacle avoidance safety, flight energy consumption can be effectively reduced. The shorter path reduces flight distance and flight time, thereby reducing the UAV's energy consumption and extending the mission execution time. It can find reasonable passage between obstacles, demonstrating its powerful path search capability in complex three-dimensional environments. It is suitable for UAV path planning tasks in complex scenarios such as urban canyons and mountain jungles.

[0093] The optimal fitness values ​​obtained using the above method or system are compared with those obtained using the dung beetle and gray wolf matching algorithm and the fusion of bat and Creer group algorithm. See [link to relevant documentation]. Figure 3 As shown, by comparing the curves of the optimal fitness value changing with the number of iterations for the three methods, it can be observed that: in terms of 3D flight paths, this application can plan smoother and safer paths; in terms of the optimal fitness value convergence curve, this application has a faster convergence speed and more stable convergence performance.

[0094] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for unmanned aerial vehicle (UAV) path planning, characterized in that, include: Acquire three-dimensional spatial information of the environment in which the drone is located, and construct an environmental model that includes the distribution of obstacles; Energy consumption constraints, self-performance constraints, and obstacle constraints are set during the flight of the UAV. Based on the cumulative sum of the cost functions corresponding to each constraint and the predetermined weight coefficients, the objective function for path evaluation is constructed. An improved chaotic quantum particle swarm optimization algorithm is used for path search: each candidate path is encoded as a particle in the search space, and the state of each particle is described by a probability density function. Candidate positions of particles are generated by Monte Carlo random sampling of the probability density function. When updating particle positions, the individual best position, the global best position, and the average best position are used as guiding paths, and the influence weight of each path is allocated by a balancing parameter. The fitness value of all particles is calculated using an objective function to determine the individual best position and the global best position of each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are superimposed on the particle position update process. It is determined whether the maximum number of iterations has been reached. If it has, the search result is output and the algorithm ends; otherwise, the particle state is updated iteratively, and a simulated annealing algorithm is introduced in the later stage of the iteration to perform a local fine search on the current optimal solution until the maximum number of iterations is reached, and the flight path that satisfies all constraints and optimizes the objective function value is obtained.

2. The UAV path planning method according to claim 1, characterized in that, Set energy consumption constraints during drone flight, including: Based on the relationship between aerodynamic formulas and energy consumption, an energy consumption model under displacement is constructed. Based on the energy consumption model under displacement, construct the energy consumption model of the UAV under different flight states; The flight path is divided into multiple segments, and a maximum range is set. The energy consumption of each segment is expressed as a function of the corresponding time when the UAV is in each flight state within that segment. The energy consumption of each segment is accumulated to obtain the total energy consumption. Set a maximum energy consumption threshold for the drone. If the calculated total energy consumption exceeds the maximum energy consumption threshold, then apply a penalty to the particle's fitness value.

3. The UAV path planning method according to claim 1, characterized in that, The self-performance constraints include speed constraints, acceleration constraints, and flight attitude constraints; The speed constraint is a penalty when the drone's flight speed exceeds the maximum speed limit; The acceleration constraint is a penalty term when the UAV's acceleration exceeds the maximum tolerable acceleration or is less than the maximum deceleration; The flight attitude constraint is a penalty when the pitch and roll angles of the UAV exceed the corresponding maximum angle limits.

4. The UAV path planning method according to claim 1, characterized in that, The obstacle constraints include: The obstacle constraint is a penalty term when the minimum distance from the UAV to the obstacle is less than the safe distance threshold; wherein, the calculation of the minimum distance includes: the minimum value of the vertical distance from the current position of the UAV to the surface of each obstacle, and then subtracting the equivalent radius of the UAV; The calculation of the safe distance threshold includes: multiplying the maximum diameter of the drone by a first safety consideration coefficient to obtain the physical size safety margin; Multiply the minimum reliable detection distance by the second security consideration coefficient to obtain the perception security margin; Multiply the minimum turning radius of the drone by the third safety consideration factor to obtain the maneuver safety margin; The safety distance threshold is obtained by summing the physical size safety margin, the perception safety margin, the maneuver safety margin, and the accuracy error.

5. The UAV path planning method according to claim 1, characterized in that, The cost function corresponding to each constraint is used to quantify the energy consumption constraint, self-performance constraint and obstacle constraint into a comparable path evaluation value, normalize each constraint, and impose corresponding penalties on candidate paths that violate the constraints, so as to guide the particle swarm to search for low-cost regions that satisfy all constraints.

6. The UAV path planning method according to claim 1, characterized in that, When updating particle positions, the following is also included: Generate a first random number that is uniformly distributed between 0 and 1. This first random number is used to perform a weighted summation of the current particle's individual best position and the population's global best position to obtain a weighted combined position. A second random number is generated that is uniformly distributed between 0 and 1. The distance weighting factor is obtained by taking the natural logarithm of the second random number and calculating its absolute value. The distance weighting factor is used to calculate the distance between the current position of the particle and the weighted combination position. A third random number is generated that is uniformly distributed between zero and one. When the third random number is less than 0.5, the candidate new position is centered on the weighted combination position, and a random offset whose amplitude is jointly controlled by the balance parameter and the distance weight factor is superimposed in the direction of the distance from the current position of the particle to the weighted combination position. When the third random number is greater than or equal to 0.5, the candidate new position is centered on the weighted combination position, and a random offset whose amplitude is jointly controlled by the balance parameter and the distance weight factor is superimposed in the direction of the distance from the average optimal position to the weighted combination position.

7. The UAV path planning method according to claim 1, characterized in that, The process of introducing chaotic sequences to generate chaotic perturbation offsets and superimposing them on the particle position update process includes: The generation rule for the chaotic sequence is as follows: when the current value of the chaotic sequence is less than 0.5, the next value is twice the current value of the chaotic sequence; when the current value of the chaotic sequence is greater than or equal to 0.5, the next value is the difference between 2 multiplied by 1 and the current value of the chaotic sequence. The chaotic perturbation amplitude is calculated based on the ratio of the current iteration number to the maximum iteration number. The chaotic perturbation amplitude decreases linearly from a preset initial value with the number of iterations, and approaches zero at the end of the iteration. Multiply the current value of the chaotic sequence by the difference in the search boundary range, and then multiply by the chaotic perturbation amplitude of the current iteration to obtain the chaotic perturbation offset.

8. The UAV path planning method according to claim 7, characterized in that, Introducing chaotic sequences to generate chaotic perturbation offsets and superimposing them on the particle position update process also includes: All particles in the population are divided into exploration and development groups according to their quantity ratio; The positions of the two groups of particles are updated independently along each coordinate axis in three-dimensional space. Each group of particles uses the current best position of the individual particle as a guiding reference, and generates candidate new positions by superimposing quantum search offset and chaotic perturbation offset. The quantum search offset is determined by multiplying the balance parameter by the distance between the current position of the particle and the average best position, and then by the distance weighting factor. The exploration group performs a global search by superimposing a chaotic perturbation offset of full magnitude, while the development group performs a local search by superimposing a chaotic perturbation offset of reduced magnitude.

9. The UAV path planning method according to claim 1, characterized in that, In the later stages of iteration, simulated annealing is introduced to perform a local fine search for the current optimal solution, including: An elite strategy is adopted to select the particles with the best fitness from the current population as elite individuals after each iteration; A new candidate position is generated by superimposing a random perturbation term on the current position of the elite individual. The random perturbation term follows a standard normal distribution, and the perturbation amplitude is controlled by the current annealing temperature. The annealing temperature decreases linearly from a preset initial value to zero with the number of iterations. Calculate the fitness value of the candidate new position. When the fitness of the candidate new position is better than that of the original elite individual position, the candidate new position is accepted unconditionally and the elite individual is updated. When the fitness of the candidate new position is worse than that of the original elite individual position, the candidate new position is accepted with a preset probability. The preset probability is determined by a negative exponential function of the fitness difference divided by the current annealing temperature. The simulated annealing process is continuously iterated until the annealing temperature drops to the preset termination temperature. The elite individual that is finally retained is then output as the global optimal solution.

10. A UAV path planning system, characterized in that, include: The environmental perception module is used to acquire three-dimensional spatial information of the environment in which the UAV is located and to construct an environmental model that includes the distribution of obstacles. The path constraint construction module is used to set energy consumption constraints, self-performance constraints, and obstacle constraints during the flight of the UAV. Based on the cumulative sum of the cost function corresponding to each constraint and the predetermined weight coefficient, the objective function for path evaluation is constructed. The path search module encodes each candidate path as a particle in the search space, describing the state of each particle with a probability density function. Candidate positions for the particles are generated by Monte Carlo random sampling of the probability density function. During particle position updates, the individual best position, global best position, and average best position are used as guiding paths, and the influence weights of each path are allocated through a balancing parameter. The fitness values ​​of all particles are calculated to determine the individual best position and global best position for each particle. A chaotic sequence is introduced to generate chaotic perturbation offsets, which are then superimposed on the particle position update process. The module determines whether the maximum number of iterations has been reached. If so, the search result is output and the algorithm terminates; otherwise, the particle state continues to be iterated and updated. In the later stages of iteration, a simulated annealing algorithm is introduced to perform a local refinement search on the current optimal solution until the maximum number of iterations is reached, yielding a flight path that satisfies all constraints and optimizes the objective function value.

Images