Unmanned aerial vehicle complex scene path planning method based on multi-objective optimization method

Through the improved NSGA-III multi-objective optimization method and Hopfiled neural network, combined with the breadth priority method, the problems of insufficient single-objective optimization and low computing efficiency in drone path planning are solved, and efficient multi-objective optimization in complex scenarios are achieved and path quality improvement.

CN119088056APending Publication Date: 2024-12-06ZHEJIANG SCI-TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411199404.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing UAV path planning methods have problems such as insufficient single-objective optimization, low computing efficiency, local minimum traps and high path randomness in complex scenarios, especially in multi-objective optimization.

Method used

The improved NSGA-III multi-objective optimization method combined with Hopfiled neural network and breadth-first method is adopted to generate the initial path through iterative update of grid numbering and energy state, and multi-objective optimization is used to introduce path smoothness costs to improve paths. quality.

Benefits of technology

It improves the success rate and multi-objective optimization capabilities of drone path planning in complex scenarios, reduces path randomness, enhances computing efficiency, and meets multiple task requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088056A_ABST
    Figure CN119088056A_ABST
Patent Text Reader

Abstract

The invention discloses an unmanned aerial vehicle complex scene path planning method based on a multi-objective optimization method. The method comprises the following steps: firstly, establishing a grid map of a working environment of an unmanned aerial vehicle, performing map segmentation, then calculating a collision-free shortest initial path by applying a Hopfield neural network, then performing path point exploration and expansion to obtain a plurality of random paths as an initial population, and then performing multi-target optimization calculation on the plurality of paths by applying an NSGA-III method to obtain a plurality of optimal paths; and finally, carrying out smoothing processing on the obtained planned path by adopting a cubic B-spline curve. By combining the Hopfield neural network method, the problem of low operation efficiency of the traditional NSGA-III method is solved, the operation efficiency of the multi-target optimization method is improved, and the multi-target optimization capability of unmanned aerial vehicle path generation is improved; by providing a Gaussian normal random number distribution strategy, the population number of the NSGA-III method is expanded, the population diversity is effectively improved, and the problem of falling into a local minimum value is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a UAV path planning method, and in particular to a UAV complex scene path planning method based on an improved NSGA-III multi-objective optimization method. Background Art

[0002] With the progress and development of science and technology, the application of quadcopter drones in human life and production is becoming more and more extensive. However, complex obstacle scenes formed by multiple obstacles of different shapes, positions and sizes or complex and rugged terrain are the main obstacles affecting the safety of drone operations. Path planning, as a key technology of drones, is the basis for them to perform various required tasks and realize autonomous navigation planning. Its purpose is to quickly find a path from the starting point to the end point in the face of various complex obstacle scenes. This path can meet the needs of various mission objectives while meeting the requirements of collision-free. At present, the mainstream methods in the field of drone path planning include A-star method, Dijkstra method, RRT method, RRT-star method, PRM method, artificial potential field method, ant colony method, particle swarm method, etc. Among them, A-star method and Dijkstra method, as variants of greedy methods, their computational efficiency will drop rapidly as the map size increases. Intelligent methods such as artificial potential field method, ant colony method and particle swarm method are easily disturbed by environmental factors and easily fall into local minimum values, and there are problems of large amount of calculation and poor real-time performance. As methods based on random sampling points, the RRT method, RRT-star method and PRM method have the disadvantages of too high path randomness, requiring a large number of iterations to obtain a better path, and too long running time. At the same time, the above mainstream methods only consider a single or at most two goals for optimization in the field of UAV path planning technology. Faced with the various mission requirements today, there is a disadvantage that they cannot meet multiple requirements and optimize them at the same time.

[0003] The NSGA-Ⅲ method is an improved method based on the genetic algorithm (GA). This method has outstanding capabilities in multi-dimensional target optimization. In terms of path planning, it can simultaneously optimize multiple targets such as path length, path smoothness, and obstacle avoidance. However, since the genetic method itself requires a large number of initial populations, it is necessary to randomly sample the map path points. When there are narrow passages or concave obstacles in the environment, the success rate of the method is low, and it is very easy to fall into the local minimum, which is not suitable for real-time path planning. The Hopfiled neural network path planning method is a planning method proposed by using the energy convergence and stability characteristics of the Hopfiled neural network. This method overcomes the good convergence characteristics of other methods in high-dimensional space, has a high computing rate, and obtains the shortest path without collision in space. However, the shortest path advantage obtained by the Hopfiled method cannot be well combined with the kinematic or dynamic constraints of the drone, and can only generate the shortest path but cannot optimize other objectives, which has limitations in the field of multi-objective optimization. Summary of the invention

[0004] In view of the various problems existing in the current mainstream methods in themselves and in multi-objective optimization, as well as the problems existing in the NSGA-III multi-objective optimization method, the present invention proposes a complex scene path planning method for unmanned aerial vehicles based on an improved NSGA-III multi-objective optimization method, which overcomes the defects of the above-mentioned prior art, improves the planning success rate of the method in complex scenes, reduces the randomness of the path, and improves the multi-objective optimization capability, so that the planned path can meet various task requirements while being collision-free.

[0005] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0006] Step 1: Get a work environment map;

[0007] Step 2: After rasterizing the work environment map Map1, an initial grid map is obtained, as well as a starting point grid and an end point grid;

[0008] Step 3: Starting from the starting grid to the ending grid, all the flyable grids in the initial grid map are numbered using the breadth-first method to obtain a grid numbering map;

[0009] Step 4: Generate a collision-free path on the grid number map that meets multi-objective optimization using the improved NSGA-Ⅲ method;

[0010] Step 5: After optimizing the path generated in step 4, the planned path of the UAV is obtained.

[0011] The step 3 is specifically as follows:

[0012] Step 31: Fill the values ​​of the flyable grids, grids where obstacles are located, and boundary grids in the initial grid map, wherein the values ​​of the flyable grids are the same, and the values ​​of the grids where obstacles are located and boundary grids are the same, thereby completing the initialization of the initial grid map;

[0013] Step 32: Start numbering from the starting grid, and use the breadth-first method to sequentially number the grids in eight directions from the starting grid until the end grid, to obtain a grid numbering map.

[0014] The step 4 is specifically as follows:

[0015] Step 41: using the Hopfiled neural network to add energy to the grid of the grid number map and iteratively update the energy state to obtain a grid map containing the energy state;

[0016] Step 42: Use the backtracking method to find a path with the maximum energy gradient from the end point to the starting point in the grid map containing the energy state, and use this path as the collision-free initial path P init ;

[0017] Step 43: Compare the collision-free initial path P with the path length init After equal division, obtain the coordinates of N path points;

[0018] Step 44: randomly obtaining num candidate path points that obey a two-dimensional Gaussian normal distribution near the coordinates of the N path points, a total of N×num candidate path points, thereby forming N candidate areas;

[0019] Step 45: Use the NSGA-III method to select the path points in the N candidate areas and perform multi-objective iterative optimization to obtain the optimal N path points. Connect the starting grid, the optimal N path points and the end grid to obtain a path P that has no collision and can meet the multi-objective optimization. optimize .

[0020] In step 41, the specific formula for adding energy to the grid of the grid number map and iteratively updating the energy state using the Hopfiled neural network is as follows:

[0021]

[0022]

[0023] Among them, X i (k+1) represents the energy state of grid i at the k+1th iteration, X j (k) and X j(k+1) are the energy states of grid j at the kth and k+1th iterations, respectively, and f(X) represents the total energy input of grid i; D i is the grid energy coefficient, the obstacle grid and boundary grid are both 0, and the other grids are 1; I represents the initial input energy of the starting grid, A represents the energy attenuation coefficient, ω ij represents the weight between grid i and grid j in the Hopfield neural network, k is the number of iterations, NE i Represents the set of adjacent grids of grid i.

[0024] In step 45, the NSGA-III method randomly selects a candidate path point in each of the N candidate areas to obtain an initial path with a total of N path points. t There are dim initial paths in total. After traversing and searching in the dim initial paths, the optimal path consisting of N path points is obtained.

[0025] In step 45, the fitness function adopted by the NSGA-III method includes path length, obstacle threat cost and path smoothness cost. The path length, obstacle threat cost and path smoothness cost are all used as optimization targets, thereby forming multiple optimization targets.

[0026] The formula for the path length f1 is as follows:

[0027]

[0028] Where n is the number of path points, P i,j represents the jth path point of the i-th path, P i,j+1 represents the j+1th path point of the i-th path, ||P i,j →P i,j+1 || represents the adjacent path point P i,j and P i,j+1 The Euclidean distance between .

[0029] The formula of the obstacle threat cost f2 is as follows:

[0030]

[0031]

[0032] Where n is the number of path points, m is the number of obstacles, and T k (P i,j →P i,j+1 ) is about obstacles and local path P i,j →P i,j+1 The distance between k A piecewise function of , d is the obstacle radius, r kis the threat radius of the kth obstacle, γ c () is the penalty coefficient.

[0033] The formula of the path smoothness cost f3 is as follows:

[0034]

[0035] Where n is the number of path points, P i,j represents the jth path point of the i-th path, P i,j+1 represents the j+1th path point of the i-th path, P i,j-1 represents the j-1th path point of the i-th path, ||(P i,j -P i,j+1 )+(P i,j-1 -P i,j )|| 2 Represents the jth path point P on the i-th path i,j The combined force.

[0036] The step 5 is specifically as follows:

[0037] The path P generated in step 4 is optimized using the cubic B-spline optimization method. optimize After smoothing, the plane motion curve is obtained, and then the plane motion curve is linearly interpolated in height according to the height of the starting point and the end point to obtain the planned path of the UAV.

[0038] The beneficial effects of the present invention are:

[0039] This paper proposes a complex scene path planning method for unmanned aerial vehicles based on an improved NSGA-III multi-objective optimization method. The paper improves the existing mainstream methods and NSGA-III methods from the following four aspects:

[0040] 1) In view of the problem that the existing mainstream methods can only achieve single-objective optimization and cannot perform multi-objective optimization as the multi-task requirements change, an improved NSGA-III multi-objective optimization method is proposed. As a new generation of methods in the field of multi-objective optimization, the NSGA-III method has excellent multi-dimensional objective optimization characteristics and strong optimization capabilities.

[0041] 2) In order to solve the problem that the initial population of the NSGA-III method selects path points by random sampling, which leads to low success rate, low computational efficiency and easy falling into local minimum, a Hopfield neural network path planning method is proposed to generate the initial path, which avoids the problem of unsuccessful path selection and falling into local minimum when facing narrow obstacles, reduces the computational time of the method and improves the computational efficiency.

[0042] 3) In view of the problem that the NSGA-III method requires a large number of initial populations, but the path generated by the Hopfield method only generates a fixed path in the same scenario, the Gaussian normal random number distribution method is introduced to expand new path points near the initial path generated by the Hopfield method, providing a rich population sample for the NSGA-III method, avoiding the method from falling into a local optimal state, and improving the optimization of the method in other aspects.

[0043] 4) Path smoothness is introduced as one of the multi-objective optimization functions to improve the success rate of the UAV flight path meeting the trajectory tracking requirements and improve the efficiency of the next step of cubic B-spline smoothing optimization.

[0044] Practice has proved that the method provided by the present invention can be applied to UAV path planning in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the method of the present invention;

[0046] Figure 2 is a flow chart of the NSGA-III method part of the present invention;

[0047] Figure 3 is a three-dimensional model diagram of an example drone of the present invention;

[0048] Figure 4 It is a simulation process diagram of the UAV path planning method based on the improved NSGA-III method in the present invention in the first environment;

[0049] Figure 5 It is a simulation process diagram of the UAV path planning method based on the improved NSGA-III method in the present invention in the second environment;

[0050] Figure 6 It is a comparison chart of the path lengths of the UAV path planning methods of the improved A* method, the improved RRT* method and the improved NSGA-III method of the present invention in two environments. DETAILED DESCRIPTION

[0051] The invention will be further described below with reference to the accompanying drawings and specific implementation examples.

[0052] like Figure 1 As shown, a UAV path planning method based on the improved NSGA-Ⅲ method includes the following steps:

[0053] Step 1: Obtain a work environment map; for example, collect detailed information about the work environment through the existing UAV work environment information and by equipping the UAV with a lidar sensor, and then use information fusion and SLAM mapping technology to build a work environment map, recorded as Map1.

[0054] Step 2: Take a quad-rotor drone as an example. Figure 3 As shown, including but not limited to such drones. Set the take-off position of the drone as the starting point QS, and the mission destination as the end point QE; after rasterizing the working environment map Map1, obtain the initial grid map, as well as the starting point grid and the end point grid; specifically, divide the working environment map Map1 into n×m grid maps according to the set drone flight step S, that is, the coordinate position U(i, j) represents the grid located in the i-th row and j-th column of the grid map. Mark the starting point QS coordinates as grid U(xS, yS), and mark the end point QE coordinates as grid U(xE, yE), where xS and yS represent the horizontal and vertical coordinates of the starting point QS on the grid map, and xE and yE represent the horizontal and vertical coordinates of the end point QE on the grid map.

[0055] Step 3: Starting from the starting grid to the ending grid, all the flyable grids in the initial grid map are numbered using the breadth-first method to obtain a grid numbering map;

[0056] Step 3 is as follows:

[0057] Step 31: Fill the values ​​of the flyable grids, grids where obstacles are located, and boundary grids in the initial grid map. The values ​​of the flyable grids are the same, recorded as 255, and the values ​​of the grids where obstacles are located and boundary grids are the same, recorded as 0, completing the initialization of the initial grid map. Figure 4 , Figure 5 and Figure 6 In the figure, the black part is the obstacle area, and the white part is the flyable area;

[0058] Step 32: Start numbering from the starting grid, number it as 1, and use the breadth-first method to sequentially number the grids in eight directions from the starting grid until the end grid, to obtain a grid numbering map.

[0059] Step 4: Generate a collision-free path on the grid number map that meets multi-objective optimization using the improved NSGA-Ⅲ method;

[0060] Step 4 is as follows:

[0061] Step 41: using the Hopfiled neural network to add energy to the grid of the grid number map and iteratively update the energy state until the energy state reaches stability and then terminates the state update to obtain a grid map Map3 containing the energy state;

[0062] In step 41, the specific formula for adding energy to the grid of the grid number map and iteratively updating the energy state using the Hopfield neural network is as follows:

[0063]

[0064]

[0065] Among them, X i (k+1) represents the energy state of grid i at the k+1th iteration, X j (k) and X j (k+1) are the energy states of grid j at the kth and k+1th iterations, respectively, and f(X) represents the total energy input of grid i; D i is the grid energy coefficient, the obstacle grid and boundary grid are both 0, and the other grids are 1; I represents the initial input energy of the starting grid, A represents the energy attenuation coefficient, ω ij represents the weight between grid i and grid j in the Hopfield neural network, k is the number of iterations, NE i Represents the set of adjacent grids of grid i.

[0066] Step 42: Use the backtracking method to find a path with the maximum energy gradient from the end point to the starting point in the grid map containing the energy state, and use this path as the collision-free initial path P init ,like Figure 4 (a) and Figure 5 As shown in (a);

[0067] Step 43: Compare the collision-free initial path P with the path length init After equal division, the coordinates of N path points are obtained; specifically, according to the set equal division step length T and the initial path P init The length of the path P is obtained by calculating the number of equally divided points N, where N is an integer. init Take the upper point and get the coordinates of N path points;

[0068] Step 44: randomly obtain num candidate path points that obey a two-dimensional Gaussian normal distribution near the coordinates of the N path points, a total of N×num candidate path points, thereby forming N candidate areas, such as Figure 4 (b) and Figure 5 as shown in (b);

[0069] Step 45: Use the NSGA-III method to select the path points in the N candidate areas and perform multi-objective iterative optimization to obtain the optimal N path points. Connect the starting grid, the optimal N path points and the end grid to obtain a path P that has no collision and can meet the multi-objective optimization. optimize ,like Figure 4 (c) and Figure 5 as shown in (c).

[0070] like Figure 2 As shown, step 45 is specifically as follows:

[0071] Step 451: Initialize population P t , where the number of individuals is dim. The individual consists of N genes, and the individual is initialized. Each individual s in the initial population i =(x i ,y i ), where i = (1,…,dim). i ,y i Represents the coordinates in the exploration scene space, randomly selects a candidate path point in each of the N candidate areas, and obtains an initial path with a total of N path points. The population P t There are dim paths in total;

[0072] Step 452: Calculate the fitness function of each path; the fitness function includes the path length f1, the obstacle threat cost f2, and the path smoothness cost f3. All three costs are used as optimization targets. The formula is as follows:

[0073]

[0074]

[0075]

[0076]

[0077] Where n is the number of path points, m is the number of obstacles, ||P i,j →P i,j+1 || represents the adjacent path point P i,j and P i,j+1 The Euclidean distance between them, k refers to the kth obstacle, T k (P i,j →P i,j+1 ) is about obstacles and local path P i,j →P i,j+1 The distance between k A piecewise function of , d is the obstacle radius, r k is the threat radius of the kth obstacle, γ c is the penalty coefficient, P i,j represents the jth path point of the i-th path, P i,j+1 represents the j+1th path point of the i-th path, P i,j-1 represents the j-1th path point of the i-th path, ||(P i,j -P i,j+1 )+(P i,j-1 -P i,j )|| 2Represents the jth path point P on the i-th path i,j The smaller the resultant force, the higher the smoothness.

[0078] Step 453: For population P t According to the fixed function value, selection, crossover and sorting operations are performed to obtain a new generation of population Q t , merged population P t , Q t Get R t .

[0079] For the combined population R t Stratification and sorting based on dominance relationships. i and j The dominance relationship between them is defined as: for all objective functions, f k (s i )≤f k (s j ), k = {1, 2, ..., q}, there exists l∈k such that f l (s i ) <f l (s j ), all individuals can be divided into different levels based on the dominance relationship to obtain a hierarchy. Individuals in the same hierarchy are sorted based on the reference point. The first dim individuals are taken to form a new generation of population P t ;

[0080] Step 454: Repeat steps 452 and 453, and terminate the iteration when the number of iterations reaches the maximum number of iterations Maxiter, to obtain the optimal Pareto solution set.

[0081] Use the following formula to get the optimal solution and output the optimal path;

[0082]

[0083] Among them, Kq represents the weight coefficient of each function, q represents the sequence number of the optimization target, and f q is f1-f3.

[0084] Step 5: After optimizing the path generated in step 4, the planned path of the UAV is obtained, and the UAV flies on the working environment map according to the planned path.

[0085] Step 5 is as follows:

[0086] The path P generated in step 4 is optimized using the cubic B-spline optimization method. optimizeAfter smoothing, a plane motion curve that can be used for trajectory tracking is obtained. Then, according to the height of the starting point and the end point, the plane motion curve is linearly interpolated in height (as a three-dimensional height change) to obtain the planned path of the UAV, such as Figure 4 (d) and Figure 5 as shown in (d).

[0087] In order to verify the effectiveness and feasibility of the UAV path planning method based on the improved NSGA-III multi-objective optimization method proposed in this paper, simulation was carried out on MatLab R2020b, in the hardware environment of i5-12490F, processor 3.00GhzCPU, and memory 16.0GB. The maps are 500×500 and 500×300 respectively.

[0088] In order to verify that the UAV path planning method based on the improved NSGA-III multi-objective optimization method proposed in the present invention has advantages over the current mainstream methods, the improved A* method and the improved RRT* method are used to conduct simulation experiments with the method of the present invention under the same map. Table 1 summarizes the planning path lengths and operation times of these methods in different environments. Figure 6 (a) and Figure 6 (b) shows the simulation paths of the three methods.

[0089] Table 1 is a comparative analysis of the experimental results of the improved NSGA-Ⅲ method, the improved A* method, and the improved RRT* method.

[0090]

[0091] It can be seen from Table 1 that in the face of mainstream methods in the field of UAV path planning, no matter in which environment, the improved NSGA-III method of the present invention has better operating results and stronger operating efficiency. Figure 5 Displayed in.

[0092] The present invention also proposes a complex scene path planning system for unmanned aerial vehicles based on a multi-objective optimization method, comprising:

[0093] An initial grid map generation module is used to perform map rasterization processing on the working environment map Map1 to generate an initial grid map;

[0094] A grid number map generation module is used to number all flyable grids in the initial grid map from the starting grid to the end grid using a breadth-first method to generate a grid number map;

[0095] The multi-objective optimization path generation module is used to generate a collision-free path on the grid number map that meets the multi-objective optimization requirements using the improved NSGA-Ⅲ method;

[0096] The path optimization module is used to optimize the generated path and obtain the planned path of the UAV.

[0097] An embodiment of the present invention provides a UAV complex scene path planning device based on a multi-objective optimization method, comprising one or more processors for implementing a UAV complex scene path planning method based on a multi-objective optimization method in the above embodiment.

[0098] The embodiment of the complex scene path planning device of a drone based on a multi-objective optimization method of the present invention can be applied to any device with data processing capabilities, and the device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located to read the corresponding computer program instructions in the non-volatile memory into the memory for execution. From the hardware level, it is a hardware structure diagram of any device with data processing capabilities where the complex scene path planning device of a drone based on a multi-objective optimization method of the present invention is located. In addition to the processor, memory, network interface, and non-volatile memory, the manufacturing of any device with data processing capabilities in the embodiment where the device is located is usually based on the actual function of the device with data processing capabilities, and other hardware may also be included, which will not be repeated.

[0099] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0100] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] An embodiment of the present invention also provides a readable storage medium on which a program is stored. When the program is executed by a processor, a method for complex scene path planning of a drone based on a multi-objective optimization method in the above embodiment is implemented.

[0102] The readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The readable storage medium may also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0103] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A complex scene path planning method for unmanned aerial vehicles based on a multi-objective optimization method, characterized in that: The steps include: Step 1: Get a work environment map; Step 2: After rasterizing the work environment map Map1, an initial grid map is obtained, as well as a starting point grid and an end point grid; Step 3: Starting from the starting grid to the ending grid, all the flyable grids in the initial grid map are numbered using the breadth-first method to obtain a grid numbering map; Step 4: Generate a collision-free path on the grid number map that meets multi-objective optimization using the improved NSGA-Ⅲ method; Step 5: After optimizing the path generated in step 4, the planned path of the UAV is obtained.

2. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 1 is characterized in that: The step 3 is specifically as follows: Step 31: Fill the values ​​of the flyable grids, grids where obstacles are located, and boundary grids in the initial grid map, wherein the values ​​of the flyable grids are the same, and the values ​​of the grids where obstacles are located and boundary grids are the same, thereby completing the initialization of the initial grid map; Step 32: Start numbering from the starting grid, and use the breadth-first method to sequentially number the grids in eight directions from the starting grid until the end grid, to obtain a grid numbering map.

3. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 1 is characterized in that: The step 4 is specifically as follows: Step 41: using the Hopfiled neural network to add energy to the grid of the grid number map and iteratively update the energy state to obtain a grid map containing the energy state; Step 42: Use the backtracking method to find a path with the maximum energy gradient from the end point to the starting point in the grid map containing the energy state, and use this path as the collision-free initial path P init ; Step 43: Compare the collision-free initial path P with the path length init After equal division, obtain the coordinates of N path points; Step 44: randomly obtaining num candidate path points that obey a two-dimensional Gaussian normal distribution near the coordinates of the N path points, a total of N×num candidate path points, thereby forming N candidate areas; Step 45: Use the NSGA-III method to select the path points in the N candidate areas and perform multi-objective iterative optimization to obtain the optimal N path points. Connect the starting grid, the optimal N path points and the end grid to obtain a path P that has no collision and can meet the multi-objective optimization. optimize .

4. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 3 is characterized in that: In step 41, the specific formula for adding energy to the grid of the grid number map and iteratively updating the energy state using the Hopfiled neural network is as follows: Among them, X i (k+1) represents the energy state of grid i at the k+1th iteration, X j (k) and X j (k+1) are the energy states of grid j at the kth and k+1th iterations, respectively, and f(X) represents the total energy input of grid i; D i is the grid energy coefficient, the obstacle grid and boundary grid are both 0, and the other grids are 1; I represents the initial input energy of the starting grid, A represents the energy attenuation coefficient, ω ij represents the weight between grid i and grid j in the Hopfield neural network, k is the number of iterations, NE i Represents the set of adjacent grids of grid i.

5. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 3 is characterized in that: In step 45, the NSGA-III method randomly selects a candidate path point in each of the N candidate areas to obtain an initial path with a total of N path points. t There are dim initial paths in total. After traversing and searching in the dim initial paths, the optimal path consisting of N path points is obtained.

6. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 3 is characterized in that: In step 45, the fitness function adopted by the NSGA-III method includes path length, obstacle threat cost and path smoothness cost. The path length, obstacle threat cost and path smoothness cost are all used as optimization targets, thereby forming multiple optimization targets.

7. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 6 is characterized in that: The formula for the path length f1 is as follows: Where n is the number of path points, P i,j represents the jth path point of the i-th path, P i,j+1 represents the j+1th path point of the i-th path, ‖P i,j →P i,j+1 ‖ represents the adjacent path point P i,j and P i,j+1 The Euclidean distance between .

8. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 6 is characterized in that: The formula of the obstacle threat cost f2 is as follows: Where n is the number of path points, m is the number of obstacles, and T k (P i,j →P i,j+1 ) is about obstacles and local path P i,j →P i,j+1 The distance between k A piecewise function, d is the obstacle radius, r k is the threat radius of the kth obstacle, γ c () is the penalty coefficient.

9. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 6 is characterized in that: The formula of the path smoothness cost f3 is as follows: Where n is the number of path points, P i,j represents the jth path point of the i-th path, P i,j+1 represents the j+1th path point of the i-th path, P i,j-1 represents the j-1th path point of the i-th path, ‖(P i,j -P i,j+1 )+(P i,j-1 -P i,j )‖ 2 Represents the jth path point P on the i-th path i,j The combined force.

10. The method for complex scene path planning of unmanned aerial vehicle based on multi-objective optimization method according to claim 6, characterized in that: The step 5 is specifically as follows: The path P generated in step 4 is optimized using the cubic B-spline optimization method. optimize After smoothing, the plane motion curve is obtained, and then the plane motion curve is linearly interpolated in height according to the height of the starting point and the end point to obtain the planned path of the UAV.