Unmanned aerial vehicle mountainous region path planning system based on adaptive basis vector evolutionary algorithm

Abstract

The invention discloses an unmanned aerial vehicle mountain path planning system based on an adaptive basis vector evolutionary algorithm, and relates to the field of unmanned aerial vehicle path planning. A continuous three-dimensional mountain landform is generated by adopting multi-peak Gaussian function superposition, then a comprehensive constraint model is constructed, the path length, the flight height and the smoothness cost are quantized, candidate paths are coded into a matrix through an adaptive basis vector evolutionary algorithm, an evolutionary population is constructed through replacement sampling, and the three-dimensional mountain landform is obtained. Four mechanisms of adaptive basis vector, historical memory, attenuation global disturbance and adaptive difference step length are fused in iteration, and global exploration and local development are effectively balanced, so that a smooth and safe optimal flight path with the minimum total cost is quickly converged; according to the unmanned aerial vehicle mountain path planning system based on the adaptive basis vector evolutionary algorithm, efficient and reliable technical guarantee is provided for autonomous and safe flight of the unmanned aerial vehicle in a complex mountain environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning, and in particular to a UAV mountain path planning system based on an adaptive basis vector evolution algorithm. Background Technology

[0002] With the widespread application of drones in tasks such as mountain search and rescue, geological mapping, power line inspection, and forest fire prevention, their autonomous flight capability in complex mountain environments has become a key factor restricting the success or failure of these missions. Mountain environments are characterized by dramatic terrain undulations, intersecting ridges and valleys, numerous saddles, and spatially continuous differentiable features, posing extremely high challenges to three-dimensional path planning. In existing technologies, A... Deterministic algorithms such as A-Star Search Algorithm and Rapidly-exploring Random Tree (RRT) suffer from problems such as high computational cost and severe path jaggedness. Methods such as Particle Swarm Optimization and Classical Differential Evolution are prone to premature convergence in high-dimensional path optimization, resulting in slow convergence speed, low accuracy, and difficulty in effectively handling kinematic constraints such as yaw and pitch angles. Adaptive algorithms such as SHADE (Adaptive Differential Evolution Algorithm Based on Success History) and L-SHADE (SHADE Algorithm with Linear Population Reduction) that have emerged in recent years, although they incorporate historical parameter memory, still adopt a single basis vector strategy, making it difficult to achieve a dynamic balance between global exploration and local development. This leads to poor smoothness of the planned path, insufficient safety margin, and high overall cost, making it difficult to meet the actual needs of UAVs for safe and efficient flight in complex and continuous mountainous environments. Summary of the Invention

[0003] The purpose of this invention is to provide a UAV mountain path planning system based on an adaptive basis vector evolutionary algorithm. It constructs a continuously differentiable 3D mountain environment using a Gaussian superposition strategy, proposes a comprehensive cost model including constraints on path length, flight altitude, yaw angle, and pitch angle, and designs a matrix evolutionary algorithm that integrates diagonal and weighted dual-track basis vectors, historical path memory, attenuation of global disturbances, and adaptive differential learning. A complete closed-loop system of "mountain environment generation—comprehensive constraint modeling—high-dimensional matrix evolutionary optimization" is constructed, achieving full automation from complex terrain input to safe, smooth, and low-cost optimal flight trajectory output. This system effectively overcomes the shortcomings of traditional evolutionary algorithms in high-dimensional, strongly constrained spaces, such as premature convergence, low search accuracy, and poor path flyability, significantly improving optimization efficiency and solution quality, and providing efficient and reliable technical support for the autonomous and safe flight of UAVs in complex mountain environments.

[0004] To achieve the above objectives, this invention provides an unmanned aerial vehicle (UAV) mountain path planning system based on an adaptive basis vector evolution algorithm, including a Gaussian superposition mountain environment generation module, an UAV path comprehensive constraint model module, and an adaptive basis vector evolution algorithm module. The Gaussian superposition mountain environment generation module uses a multi-peak Gaussian function superposition strategy to construct a three-dimensional continuous mountain environment. It linearly superimposes several independent Gaussian peaks to achieve efficient parametric modeling of complex natural landforms. By adjusting parameters such as the number of peaks, spatial distribution, and morphology, it can quickly generate continuous, differentiable, smooth, and complexity-controllable mountain terrain and output terrain elevation information to the UAV path integrated constraint model module. The UAV path comprehensive constraint model module takes minimizing the total length of the flight path from the take-off point to the landing point as its primary optimization objective. This module organically integrates different weights for path length cost, UAV flight altitude cost, and overall path smoothness cost, and constructs a complete mathematical optimization model by combining relevant definitions and penalty terms. The mathematical optimization model and objective function are then transmitted to the adaptive basis vector evolution algorithm module as the evaluation criteria for algorithm iterative optimization. The adaptive basis vector evolution algorithm module receives the objective function from the UAV path integration constraint model module as the optimization objective. It encodes candidate paths into a matrix and generates an initial population through sampling with replacement. During the iterative optimization process, the algorithm integrates mechanisms such as adaptive basis vectors, historical memory, and decay perturbation to dynamically adjust the search direction. Whenever the algorithm generates a new path individual, it needs to send the coordinate information of the path back to the UAV path integration constraint model module, call its objective function to evaluate the fitness, and obtain the quantified cost value as the basis for optimization. Finally, this module outputs the converged optimal flight path scheme that meets all constraints to the system. Therefore, the UAV mountain path planning system based on the adaptive basis vector evolution algorithm described above has the following advantages compared with the prior art: 1. For the first time, diagonal basis vectors and weighted basis vectors are adaptively fused through historical evolution paths. Combined with decay perturbation and differential learning of superior and inferior individuals, the algorithm significantly improves its ability to balance global exploration and local development in high-dimensional path optimization problems. 2. Without the need for additional constraint processing modules, it can guarantee 100% that the path will not collide with mountains, the yaw and pitch angles will be strictly limited, and the flight altitude will be within a safe range. The generated trajectory is natural and smooth, conforming to mountain ridges and valleys, greatly improving the feasibility of actual flight.

[0005] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0006] Figure 1This is a schematic diagram of the structure of an unmanned aerial vehicle (UAV) mountain path planning system based on an adaptive basis vector evolution algorithm according to the present invention. Detailed Implementation

[0007] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed when in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0008] Example like Figure 1 As shown, the present invention provides a UAV mountain path planning system based on an adaptive basis vector evolution algorithm, which includes a Gaussian superposition mountain environment generation module, a UAV path comprehensive constraint model module, and an adaptive basis vector evolution algorithm module. The Gaussian superposition mountain environment generation module uses a multi-peak Gaussian function superposition strategy to construct a three-dimensional continuous mountain environment. It linearly superimposes several independent Gaussian peaks to achieve efficient parametric modeling of complex natural landforms. By adjusting parameters such as the number of peaks, spatial distribution, and morphology, it can quickly generate continuous, differentiable, smooth, and complexity-controllable mountain terrain and output terrain elevation information to the UAV path integrated constraint model module. Traditional methods for generating 3D mountain terrain mainly include those based on height maps, geometric structures, or physical process simulations. However, these methods have certain limitations: they struggle to accurately represent complex mountain structures with multiple independent peaks and controllable morphology; the generated terrain often lacks clear terrain feature parameters; or they suffer from low computational efficiency and unintuitive adjustments. To address these shortcomings, the Gaussian stacking mountain environment generation module uses the superposition of multiple rotationally symmetric Gaussian functions to generate complex 3D mountain environments. The height at any point on the surface of the mountain environment is linearly superimposed from multiple independent Gaussian peaks, as specifically expressed as: in, Given the coordinates of any point on the surface of a mountainous environment. This represents the altitude corresponding to that point. Indicates the first The center coordinates of each mountain peak Indicates the first The elevation of the summit of each mountain peak and The mountain peaks are respectively Gaussian decay radius in the direction, The total number of mountain peaks; The Gaussian stacking mountain environment generation module accurately simulates the slope variation of real mountains through the rapid decay characteristics of the exponential function: the farther away from the center of the mountain, the more significant the height decay, thus naturally forming a smooth mountain outline that conforms to physical laws. In the actual modeling process, several single Gaussian peaks with different center coordinates, heights, and widths are first constructed independently. Then, their height fields are linearly stacked to efficiently generate complex natural mountain landforms with multiple peaks, continuous ridges, intersecting valleys, and obvious saddles. The method in this module not only ensures the continuity and second-order differentiability of the surface height function, which is convenient for subsequent path smoothing and collision detection, but also has clear physical meaning of parameters and intuitive adjustment. It can quickly generate diverse and fully reproducible three-dimensional complex terrains, providing a highly realistic and controllable test environment for subsequent path planning algorithms. The UAV path comprehensive constraint model module takes minimizing the total length of the flight path from the take-off point to the landing point as its primary optimization objective. This module organically integrates different weights for path length cost, UAV flight altitude cost, and overall path smoothness cost, and constructs a complete mathematical optimization model by combining relevant definitions and penalty terms. The mathematical optimization model and objective function are then transmitted to the adaptive basis vector evolution algorithm module as the evaluation criteria for algorithm iterative optimization. The UAV path integration constraint model module searches for an optimal, collision-free trajectory from a specified starting point to a target point in the solution space, while considering terrain obstacles and flight physics constraints, to determine feasible paths. Represented as a series of ordered path points ,in It is the starting point. The target point, each path point Boundary constraints must be satisfied. and Define the horizontal boundary range. Define the boundary range in the vertical direction; The UAV path integration constraint model module selects minimizing the UAV path length as the optimization objective. Based on the Euclidean distance formula, the maximum and minimum path lengths are defined as follows: The drone path length is determined by It consists of n path segments, where the nth The Euclidean distance of the path segment is : ; in, For the first The x-coordinates of the path points For the first The x-coordinates of the path points No. The ordinates of the path points are: For the first The y-coordinates of the path points For the first The vertical coordinates of each path point relative to the ground. For the first The vertical coordinates of each path point relative to the ground; The Euclidean distance from the starting point to the target point of the drone is defined as the minimum path length. : ; in, The x-coordinate of the target point The x-coordinate of the starting point The ordinate of the target point. The x-coordinate of the starting point Here are the vertical coordinates of the target point relative to the ground. The vertical coordinates of the starting point relative to the ground; The maximum path length is defined as the total path length from which a drone ascends vertically from its starting point to its maximum safe flight altitude, flies in a straight line directly above the target point, and then descends vertically. : ; in, It is a constant value representing the highest safe altitude that the drone is allowed to reach in that flight airspace or mission; according to , , Define drone path length penalty As shown in the following formula: ; The UAV path integration constraint model module sets the upper limit of the UAV's flight altitude relative to the terrain. With lower limit of flight altitude any path point actual flight altitude The following relationship must be satisfied: ; in, For the first The elevation of the ground surface directly below each path point is calculated in real time by the Gaussian stacking mountain model. And define high-level penalty items. for: ; in, , These represent the maximum and minimum penalty values ​​when the flight altitude violates the maximum safe altitude constraint. It is infinitely large; In highly penalized items In this process, if any path point violates the lower bound, it is directly given an infinite penalty; if it exceeds the upper bound, a gradual normalization penalty is applied. The UAV path integration constraint model module introduces dual constraints of yaw angle and pitch angle as core indicators for measuring and ensuring path smoothness. The angle between two adjacent flight segments at a waypoint in the horizontal projection plane is defined as the horizontal yaw angle. : ; in, It is an inverse cosine function. The maximum allowable yaw angle for the drone, and the horizontal vector. and Defined as: ; No. The tangent of the rate of change of altitude between two adjacent flight segments at each waypoint is defined as the vertical pitch angle. : ; in, It is the arctangent function. These are the maximum allowable vertical pitch angles for the drone; Define horizontal yaw angle penalty term Vertical pitch angle penalty for: ; ; in, , These are the maximum and minimum penalty values ​​for violating the horizontal yaw angle constraint, respectively. , These are the maximum and minimum penalty values ​​for violating the vertical pitch angle constraint, respectively. The quality of the generated UAV path is determined by the fitness function in the UAV path integration constraint model module. A unified quantitative assessment is conducted. This function comprehensively weighs multiple factors, including total path length, terrain collision risk, flight altitude rationality, and trajectory smoothness. Its mathematical expression is: ; in, , , and These represent the path length penalty, path altitude penalty, yaw angle penalty, and pitch angle penalty, respectively. This is the path length penalty coefficient. As a high penalty coefficient, and These are the penalty coefficients for the horizontal yaw angle and the vertical pitch angle, respectively. The adaptive basis vector evolution algorithm module receives the objective function from the UAV path integration constraint model module as the optimization objective. It encodes candidate paths into a matrix and generates an initial population through sampling with replacement. During the iterative optimization process, the algorithm integrates mechanisms such as adaptive basis vectors, historical memory, and decay perturbation to dynamically adjust the search direction. Whenever the algorithm generates a new path individual, it needs to send the coordinate information of the path back to the UAV path integration constraint model module, call its objective function to evaluate the fitness, and obtain the quantified cost value as the basis for optimization. Finally, this module outputs the converged optimal flight path scheme that meets all constraints to the system. The adaptive basis vector evolutionary algorithm module organizes candidate paths in matrix form, quickly generates evolutionary matrices through sampling with replacement, and feeds them into the UAV path integration constraint model module to calculate the objective function value of each path. In each iteration, it integrates four mechanisms: adaptive basis vectors, historical evolutionary path memory, decay of global disturbances, and adaptive differential step size to achieve a dynamic balance between exploration and development. Combined with strict boundary control and greedy selection, it quickly converges to a smooth, safe, and low-cost optimal flight trajectory. The adaptive basis vector evolutionary algorithm module organizes candidate paths in matrix form, generates an evolutionary matrix through sampling with replacement, and feeds it into the UAV path comprehensive constraint model module to calculate the path objective function value. Traditional UAV path planning methods are mostly based on swarm intelligence algorithms such as particle swarm optimization (PSO). In high-dimensional mountainous environments, these methods often face limitations such as slow convergence speed, weak local search ability, and susceptibility to local optima. They also struggle to simultaneously meet the requirements of smoothness, safety, and multiple constraints in path generation. To overcome these problems, during the iteration process, adaptive basis vectors, historical evolution path memory, attenuation of global disturbances, and adaptive differential step size are comprehensively utilized to achieve a dynamic balance between global exploration and local development. Combined with boundary constraint correction and greedy selection strategies, the convergence to a smooth, safe, and low-cost optimal flight trajectory is accelerated. This overcomes the problems of slow convergence and insufficient search accuracy of traditional particle swarm optimization in high-dimensional path planning, significantly improving the optimization efficiency and solution quality of path planning. The working process of the adaptive basis vector evolution algorithm module is as follows: S1. First, flatten the intermediate path points into a one-dimensional decision vector in order from front to back. Based on boundary and height constraints, a unified upper bound is given for the flattened one-dimensional variable of each path point. and the lower realm ; The adaptive basis vector evolution algorithm module flattens the intermediate path points into one-dimensional decision vectors in a forward-to-back order. : ; in, Let D be a real space. Decision vector Total dimensions and ; Based on boundary and height constraints, a unified upper bound is given for the flattened one-dimensional variable of each path point. and the lower realm : ; Each component satisfies: If the first Dimensional correspondence coordinates, then , ; If the first Dimensional correspondence coordinates, then , ; If the first Dimensional corresponding height Coordinates are then combined with the terrain elevation function. Upper and lower limits of drone flight altitude relative to terrain During initialization and evolution, through ; S2. Construct the initial candidate path matrix The process is iterated, and an evolutionary matrix is ​​constructed based on the initialized candidate path matrix after iteration. ; The initialization process of the candidate path matrix is ​​as follows: Let the population size be Then in the first In the next iteration, the candidate paths are organized in matrix form as follows: ; in, Indicates the first The decision vector of each candidate path; During initialization At that time, for each individual Each dimension Perform uniform random initialization: ; in express Independent uniform random variables on the interval; thus quickly generating an initial candidate path group covering the entire feasible solution space; and dividing each volume Substitute the fitness function into the UAV path integration constraint model module The quality of the path can be obtained by conducting an evaluation; Initialized candidate matrix after iteration Constructing a same-dimensional evolutionary matrix ; For each row Independently Randomly sample an index with replacement And order: ; This mechanism, while maintaining linear computational complexity, increases the reuse and recombination of excellent individuals, providing a rich search baseline for subsequent Alpha operators; S3. Define two evolution paths and their corresponding basis vector construction methods; The specific process of constructing basis vectors includes: The construction of diagonal basis vectors and the process are as follows: In the The next iteration, from the population Independent random sampling Individual Index Construct a diagonal matrix : ; Its diagonal vector is: It independently mixes the superior components of different individuals in each dimension, making it suitable for handling path parameters that are relatively separable in each dimension. The construction of the weighted basis vectors and the process are as follows: Record No. In the next iteration of the population, the first The fitness of each individual is The fitness is normalized to obtain the weights. : ; in, It is a monotonic transformation function. ,in This is a scaling factor, allowing the optimal path to receive greater weight. Based on this weight, construct a weighted basis vector: ; This vector statistically characterizes the centroid of the probability distribution of the current population in the path space, and is suitable for the co-evolution of inseparable path parameters; Two evolutionary paths of design and The evolution trends of the diagonal basis vectors and the weighted basis vectors are tracked separately: ; ; in The learning rate gradually decreases with iteration. ; This represents the number of fitness function evaluations performed so far. To maximize the number of evaluations allowed, new search information gradually replaces old information as iterations progress, in order to avoid premature convergence; When constructing the Alpha operator, for each evolutionary individual Adaptive basis vectors are generated by selecting one of the two evolutionary paths with a 50% probability. : ; in The weighting coefficients are dynamically adjusted with iterations, initially favoring overall population search, and later strengthening local fine-tuning. S4. Obtaining the adaptive basis vectors Then, the Alpha operator is used to evaluate the evolutionary individuals. Update and generate new candidate paths. ; To maintain global exploration capability, a random perturbation related to the search space scale is introduced for each dimension. Generate two independent random numbers and and , and These are two random numbers independently sampled from a uniform distribution of 0 to 1, where... This represents a 0-1 uniform distribution; Construct the basic perturbation vector : ; The above form ensures that the direction of the disturbance is evenly distributed in each dimension, and the magnitude is proportional to the feasible interval of the variable; To enable selective updates across different dimensions, an independent Bernoulli masking matrix is ​​introduced. ,when At that time, with probability Keeping this dimension constant, when At that time, with probability Update this dimension; thus obtaining the actual perturbation: ,in Represents element-wise product; The disturbance strength is determined by a coefficient that decreases with iteration. control, The specific formula is as follows: ; in, The initial disturbance amplitude, This is a decay factor, which maintains a large step size for global search in the early stages and automatically shrinks the step size in the later stages to facilitate local convergence. To explicitly introduce gradient-like information, a learning term based on the difference between superior and inferior individuals is introduced for each evolutionary individual. : In terms of fitness Randomly select a path from the individuals In fitness inferior Randomly select a path from the individuals Therefore, the difference learning vector is constructed. as follows: ; in, As a basis vector compensation term, it is used to correct the deviation between the basis vector and the current individual; To balance collaborative search and dimensional decoupling, two differential scaling modes are adopted: one is global uniform scaling, and the other is dimensional independent scaling, with an adaptive switching between the two with a 50% probability. The specific formula is as follows: ; in, For the first The iteration of the ... An evolving individual The differential scaling function used, This is a scalar scaling factor. The vector coefficients are generated randomly or adaptively. The former enhances the coordinated changes between the dimensions of the path, while the latter is beneficial for fine-tuning the dimensions under complex constraints. For the An evolving individual New candidate paths generated by the Alpha operator The expression is as follows: ; The updated path also inherits the adaptive basis vectors. It possesses global statistical information, global exploration capability of random perturbation terms, and local development and gradient-guided characteristics of difference learning terms; S5. To ensure that the updated path is still within the defined search space boundary, a boundary constraint repair strategy is adopted. The boundary constraint repair strategy adopts biplane projection repair, and its specific process is as follows: right Each dimension Perform boundary checks: ; in, The binary correction method, which corresponds to the dimension value of the individual before evolution, avoids physically unreachable paths caused by going out of bounds and partially retains useful information in the search direction, thus maintaining the smoothness of the algorithm's convergence behavior while ensuring feasibility. For high dimensions Based on the above projection, and combined with the aforementioned height safety interval and height penalty term, the fitness function applies additional penalties to paths that violate local height constraints, thereby achieving dual protection through hard boundary constraints and soft penalty guidance. S6. After completing the Alpha operator update and boundary repair, the new candidate paths are... The fitness of the algorithm is calculated based on the fitness function to select a strategy and set the termination condition for the adaptive basis vector evolution algorithm. Calculate new candidate paths fitness is ,in The fitness function that comprehensively considers path length, terrain collision, flight altitude, and attitude smoothness is as described above; A one-to-one retention selection strategy is adopted: ; If the new path is better under the comprehensive penalty, the original path is replaced; otherwise, the original solution is retained. Through this local competition mechanism, the fitness of the population is guaranteed to remain monotonically unchanged, thereby gradually converging towards the globally optimal UAV path. The termination condition of the algorithm is set as follows: The maximum number of iterations or function evaluations (MaxFEs) is reached. The improvement in optimal fitness is less than a given threshold for several consecutive generations; Other convergence criteria set by external task requirements; The path with the lowest fitness in the current population is selected when one of the above conditions is met as the termination condition. The corresponding path point sequence This is considered the optimal solution to the drone path planning problem in mountainous environments, and is directly sent to the drone for execution through the system.

[0009] Therefore, this invention presents an UAV mountain path planning system based on an adaptive basis vector evolutionary algorithm. It constructs a continuously differentiable three-dimensional mountain environment using a Gaussian superposition strategy, proposes a comprehensive cost model including constraints on path length, flight altitude, yaw angle, and pitch angle, and designs a matrix evolutionary algorithm that integrates diagonal and weighted dual-track basis vectors, historical path memory, attenuation of global disturbances, and adaptive differential learning. A complete closed-loop system of "mountain environment generation—comprehensive constraint modeling—high-dimensional matrix evolutionary optimization" is constructed, achieving full automation from complex terrain input to safe, smooth, and low-cost optimal flight trajectory output. This system effectively overcomes the shortcomings of traditional evolutionary algorithms in high-dimensional, strongly constrained spaces, such as premature convergence, low search accuracy, and poor path flyability, significantly improving optimization efficiency and solution quality, and providing efficient and reliable technical support for the autonomous and safe flight of UAVs in complex mountain environments.

[0010] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An unmanned aerial vehicle (UAV) mountain path planning system based on an adaptive basis vector evolution algorithm, characterized in that, The Gauss superposition mountain environment generation module, the unmanned aerial vehicle path comprehensive constraint model module and the adaptive basis vector evolution algorithm module are included. The Gauss superposition mountain environment generation module adopts a multi-peak Gaussian function superposition strategy to construct a three-dimensional continuous mountain environment, linearly superimposes a plurality of independent Gaussian mountain peaks, realizes efficient parameterized modeling of complex natural terrain, quickly generates a continuous, differentiable, smooth and complexity controllable mountain terrain by adjusting parameters such as the number, spatial distribution and shape of the mountain peaks, and outputs the terrain elevation information to the unmanned aerial vehicle path comprehensive constraint model module. The unmanned aerial vehicle path comprehensive constraint model module takes minimizing the total length of the flight path between the takeoff point and the landing point as the primary optimization target, this module gives different weights to the path length cost, the unmanned aerial vehicle flight height cost and the path overall smoothness cost and organically combines them, and then constructs a complete mathematical optimization model in combination with related definitions and penalty terms, and the mathematical optimization model and the objective function are transmitted to the adaptive basis vector evolution algorithm module as the evaluation criterion for algorithm iterative optimization. The adaptive basis vector evolution algorithm module receives the objective function from the unmanned aerial vehicle path comprehensive constraint model module as the optimization target, encodes the candidate path into a matrix, and generates an initial population through sampling with replacement, in the iterative optimization process, the algorithm combines mechanisms such as adaptive basis vector, historical memory and decay disturbance, dynamically adjusts the search direction, whenever the algorithm generates a new path individual, the coordinate information of the path needs to be sent back to the unmanned aerial vehicle path comprehensive constraint model module, the objective function is called to evaluate the fitness, and a quantitative cost value is obtained as the optimization basis, and the module finally outputs the optimal flight path scheme that meets all the constraint conditions to the system.

2. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 1, characterized in that: The Gauss superposition mountain environment generation module generates a complex three-dimensional mountain environment in a manner of superimposing a plurality of rotationally symmetric Gaussian functions, the height of any point on the ground in the mountain environment is linearly superimposed by a plurality of independent Gaussian mountain peaks, and the specific expression is: where, is the coordinate of any point on the ground in mountainous environment, is the corresponding altitude of the point, represents the central coordinate of the th mountain peak, represents the peak altitude of the th mountain peak, and are the Gaussian attenuation radii of the mountain peak in the direction, respectively, is the total number of mountain peaks.

3. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 1, characterized in that: The unmanned aerial vehicle path synthesis constraint model module is to find an optimal and collision-free trajectory from a specified starting point to a target point in a solution space, while considering terrain obstacles and flight physical constraints is represented as a series of ordered path points, i.e. wherein is a starting point, is a target point, and each path point must satisfy boundary constraints, and defines a boundary range in the horizontal direction, defines a boundary range in the vertical direction; The unmanned aerial vehicle path comprehensive constraint model module selects minimizing the path length of the unmanned aerial vehicle as the optimization target, and defines the maximum and minimum path lengths according to the Euclidean distance formula as follows: The length of the path of the drone is composed of path segments, wherein the Euclidean distance of the first path segment is : ； wherein is the horizontal coordinate of the th path point, is the horizontal coordinate of the th path point, is the vertical coordinate of the th path point, is the vertical coordinate of the th path point, is the vertical coordinate of the th path point relative to the ground, is the vertical coordinate of the th path point relative to the ground; The Euclidean distance of the UAV from the starting point to the target point is defined as the minimum path length : ； wherein is the horizontal coordinate of the target point, is the horizontal coordinate of the starting point, is the vertical coordinate of the target point, is the horizontal coordinate of the starting point, is the vertical coordinate of the target point relative to the ground, is the vertical coordinate of the starting point relative to the ground; The total path length of the UAV vertically ascending from the starting point to the highest safe flight height, flying in a straight line to the target point directly above, and then vertically descending is defined as the maximum path length : ； wherein, is a constant value representing the maximum safe altitude allowed for the UAV to reach in the flight airspace or mission; According to , , define a drone path length penalty as follows: ； The UAV path synthesis constraint model module sets an upper limit of the flight height of the UAV relative to the terrain and the lower limit of the flight height , the actual flight height of any path point satisfies the following relationship: ​ ； in, For the first The elevation of the ground surface directly below each path point is calculated in real time by the Gaussian stacking mountain model. and define a height penalty term is: ； wherein, , are the maximum and minimum penalty values, respectively, for a flight altitude violation of the maximum safety altitude constraint, is infinite. In the highly penalizing term if any of the path points violates the lower bound, an infinite penalty is directly assigned, and if it exceeds the upper bound, a progressive normalization penalty is applied. The UAV path integration constraint model module introduces dual constraints of yaw angle and pitch angle as core indicators for measuring and ensuring path smoothness. The angle between two adjacent flight segments at a waypoint in the horizontal projection plane is defined as the horizontal yaw angle. : ； wherein, is an inverse cosine function, is the maximum yaw angle allowed for the drone, horizontal direction vector and is defined as: ; No. The tangent of the rate of change of altitude between two adjacent flight segments at each waypoint is defined as the vertical pitch angle. : ； wherein, is the arctangent function, are respectively the maximum vertical pitch angles allowed for the drone; Defining a horizontal yaw angle penalty term with a vertical pitch angle penalty term is: ； ； wherein, , are the maximum and minimum penalty values for violating the horizontal yaw angle constraint, respectively; , are the maximum and minimum penalty values for violating the vertical pitch angle constraint, respectively. The pros and cons of the generated drone path are determined by a fitness function in the drone path synthesis constraint model module A unified quantitative evaluation is performed, and the function synthetically weighs multiple important factors including the total path length, terrain collision risk, flight height rationality, and trajectory smoothness, and the mathematical expression is as follows: ； wherein, , , and denote the path length penalty, the path height penalty, the yaw angle penalty and the pitch angle penalty, respectively, is a path length penalty coefficient, is a height penalty coefficient, and are the penalty coefficients for the horizontal yaw angle and the vertical pitch angle, respectively.

4. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 3, characterized in that: The working process of the adaptive basis vector evolution algorithm module is as follows: S1, flatten the intermediate waypoints in the order from front to back into a one-dimensional decision vector and give a uniform upper bound and lower bound to the one-dimensional variable of each flattened waypoint according to the boundary constraint and the height constraint S2, constructing an initialization candidate path matrix and iterating it, and constructing an evolution matrix according to the initialization candidate path matrix after iteration ; S3, define two evolution paths and their corresponding basis vector construction methods; S4, obtaining an adaptive basis vector After that, the evolutionary individual is updated by an Alpha operator to generate a new candidate path ; S5, to ensure that the updated path is still within the defined search space boundary, a boundary constraint repair strategy is adopted; S6. After completing the Alpha operator update and boundary repair, the new candidate paths are... The fitness of the algorithm is calculated based on the fitness function to select a strategy and set the termination condition for the adaptive basis vector evolution algorithm.

5. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 4, characterized in that: The specific process of the basis vector construction method in the S3 working process of the adaptive basis vector evolution algorithm module includes: The diagonal basis vector and its process are constructed as follows: In a first iteration, a population of N individuals is randomly sampled from the population At each iteration, a new population of N individuals is randomly sampled from the population At each iteration, a new population of N individuals is randomly sampled from the population At each iteration, a new population of N individuals is randomly sampled from the population At each iteration, a new population of N individuals is randomly sampled from the population At each iteration, a new population of N individuals ； The diagonal vector thereof is: ; independently mixed in each dimension with different individuals of the excellent components, suitable for processing each dimension relatively separable path parameters; The weighted basis vector and its process are constructed as follows: The number of iterations is recorded The fitness of the i-th individual in the population of the j-th iteration is The fitness is normalized to obtain the weight :​ ； wherein is a monotonic transformation function; The weighted basis vector is constructed based on the weight: ； This vector statistically describes the probability distribution centroid of the current population in the path space, and is suitable for the collaborative evolution of non-separable path parameters.

6. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 5, characterized in that: Two evolutionary paths designed in the S3 working process of the adaptive basis vector evolutionary algorithm module With , respectively, track the evolutionary trend of the diagonal basis vector and the weighted basis vector: ； ； wherein is a learning rate that gradually decays with iteration, ; the number of fitness function evaluations made so far, the maximum number of evaluations allowed, new search information gradually replaces old information as the iterations progress, to avoid premature convergence; In constructing the Alpha operator, for each evolving individual generate an adaptive basis vector from either of two evolutionary paths with a 50% probability : ； wherein is the weight coefficient dynamically adjusted with iterations, which is biased to the population global search in the early stage and to the local fine development in the later stage.

7. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 6, characterized in that: To maintain the global exploration ability in the S4 working process of the adaptive basis vector evolution algorithm module, a random disturbance related to the scale of the search space is introduced, and a random disturbance related to the scale of the search space is introduced Two independent random numbers And And , And are two random numbers independently sampled from a uniform distribution of 0 to 1, wherein Represents a uniform distribution of 0-1; Constructing a basis perturbation vector : ； The above form ensures that the perturbation direction is evenly distributed in each dimension, and the amplitude is proportional to the variable feasible interval; To perform selective update in different dimensions, an independent Bernoulli masking matrix is introduced When , with probability the dimension is kept unchanged, when , with probability the dimension is updated; thus the actual perturbation is where denotes element-wise multiplication. The perturbation strength is given by a coefficient that decays with iteration controlling, The specific formula is as follows: ； wherein, is the initial perturbation amplitude, is the damping factor, such that in the early stage a large step size global search is maintained, and in the later stage the step size is automatically shrunk to facilitate local convergence; To explicitly introduce gradient-like information, a learning term based on the difference between the better and the worse individual is introduced. For each evolutionary individual , a path is randomly selected among the individuals with fitness better than , and a path is randomly selected among the individuals with fitness worse than . The difference learning vector is then constructed as follows: ； wherein, as a basis vector compensation term for correcting a deviation between the basis vector and the current individual; To balance the collaborative search and dimension decoupling, two difference scaling modes are adopted, one is global uniform scaling, and the other is independent scaling in each dimension and adaptively switches between the two with a probability of 50%, and the specific formula is as follows: ； wherein is the th evolutionary individual of the th iteration employed differential scaling function, is a scalar scaling coefficient, is a random or adaptively generated vector coefficient, the former enhancing the coordinated variation among the dimensions of the path and the latter facilitating the fine-tuning of dimensions under complex constraints; For the first evolutionary individual , the Alpha operator generates a new candidate path expressed as follows: ； The updated path inherits the global statistical information of adaptive basis vector , global exploration ability of random disturbance term and local development and gradient guiding characteristics of differential learning term.

8. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 7, characterized in that: The boundary constraint repair strategy in the S5 working process of the adaptive basis vector evolution algorithm module adopts the bisection projection repair, and the specific process is as follows: Perform a bounds check on each dimension of the array : ； wherein, is the corresponding dimension value of the pre-evolution individual, this halving correction method not only avoids the physical unattainable path caused by boundary crossing, but also partially retains the useful information in the search direction, thereby maintaining the smoothness of the algorithm convergence behavior while ensuring the feasibility. For high dimension On the basis of the above projection, the height safety interval and the height penalty term are combined again, and the path that violates the local height constraint is additionally punished by the fitness function, so as to realize the double guarantee of hard boundary constraint and soft penalty guidance.

9. The unmanned aerial vehicle mountain path planning system based on adaptive basis vector evolution algorithm according to claim 8, characterized in that: S6 working process of adaptive basis vector evolution algorithm module calculates new candidate path the fitness of the new candidate path is wherein is the aforementioned fitness function considering path length, terrain collision, flight height and attitude smoothness The one-to-one reservation mode selection strategy is adopted: ； That is, if the new path is better in the comprehensive penalty sense, the original path is replaced; otherwise, the original solution is reserved, through the local competition mechanism, the population fitness is monotonically not increased, so as to gradually converge to the global optimal unmanned aerial vehicle path.

10. The unmanned aerial vehicle mountain path planning system based on the adaptive basis vector evolution algorithm according to claim 9, characterized in that: The algorithm termination condition of S6 in the adaptive basis vector evolution algorithm module is set as: The maximum iteration number or the function evaluation number MaxFEs is reached; The optimal fitness improvement amplitude of a plurality of continuous generations is lower than a given threshold value; Other convergence criteria set by external task requirements; When the termination condition meets one of the above conditions, the path with the minimum fitness in the current population The corresponding path point sequence That is, the optimal solution of the UAV mountain environment path planning problem is considered, and is directly issued to the UAV for execution by the system.

Images