High-stability unmanned aerial vehicle and hybrid multi-strategy unmanned aerial vehicle path planning method

Abstract

The invention relates to a high-stability unmanned aerial vehicle and a hybrid multi-strategy unmanned aerial vehicle path planning method applied to the related technical field of path planning, and aims to improve the optimization speed and precision of an algorithm through a multi-strategy improved eagle optimization algorithm and dynamic balance search and development of a nonlinear balance factor strategy, so that the path planning efficiency is improved. The probability that the algorithm is trapped in a local extremum is effectively reduced, higher optimization precision and higher convergence rate are achieved, compared with a traditional method, the shortest path can be found all the time when the unmanned aerial vehicle path planning problem is solved, a new solution is provided for unmanned aerial vehicle path planning, and the method is suitable for the unmanned aerial vehicle path planning. A new thought and a new method are provided for solving a complex optimization problem; and the pitch angles of the paddles are changed in a magnetic control mode so as to adapt to different requirements of different flight tasks on thrust, the problem that the regulation and control precision is reduced due to multiple mechanical structures is effectively avoided, and the speed and stability of the unmanned aerial vehicle reaching the terminal point are greatly improved.

Description

Technical Field

[0001] The present invention relates to a highly stable unmanned aerial vehicle and a path planning method for the unmanned aerial vehicle with hybrid multi-strategies, in particular to a highly stable unmanned aerial vehicle and a path planning method for the unmanned aerial vehicle with hybrid multi-strategies applied to the technical field related to path planning. Background Technique

[0002] In recent years, due to the advantages of strong mobility and flexibility, unmanned aerial vehicles have been increasingly used in various fields such as military and civilian applications, and path planning is a prerequisite for the smooth autonomous execution of tasks by unmanned aerial vehicles. Path planning mainly refers to finding a path from the starting point to the target point while avoiding obstacles under the condition of satisfying various constraints of the unmanned aerial vehicle. In current research, common algorithms include artificial potential field method, heuristic search algorithm, neural network algorithm, and various swarm intelligence algorithms, including ant colony algorithm, particle swarm algorithm, bee colony algorithm, etc.

[0003] In recent years, with the gradual adoption of intelligent systems for management and control in all walks of life, the system complexity has increased exponentially, and traditional optimization algorithms have been difficult to meet such complex and diverse requirements. Since meta-heuristic algorithms do not depend on the specific characteristics and mathematical models of the problem, but continuously approach the global optimal solution through iterative search, they have characteristics such as self-adaptability, high efficiency, and strong robustness, and have received much attention in recent years. Among them, the Aquila Optimizer (AO), a meta-heuristic optimization algorithm newly proposed in recent years, has received extensive attention due to its unique search mechanism and good optimization performance.

[0004] However, when dealing with complex optimization problems, the original AO often faces problems such as being prone to falling into local optimal solutions and having poor convergence speed. To overcome these limitations, some scholars have also tried to improve it. Li Yamei, Meng Sibo, Chen Xuelian, etc. adopted the mirror simplex method to enhance the diversity of the population in "Multi-Strategy Improved Aquila Optimization Algorithm and Its Application" [J]. Application Research of Computers, 2023, 40(05): 1352-1359, and incorporated a social free foraging strategy in its third stage to reduce the constraint on the average population position. Finally, a step-by-step strategy was introduced by referring to the artificial water algorithm to accelerate the population convergence speed. Wu Suqian, Yan Jianguo, Yang Bin, et al. "Multi-Strategy Improved Aquila Optimization Algorithm and Its Application in Path Planning" [J]. Journal of Computer Applications, 2024, cscd. First, the population was initialized by combining the Sobol sequence. Secondly, the within-species mutual assistance and optimization strategy were introduced to avoid optimization stagnation, and the golden sine operator and the idea based on particle swarm self-learning and social learning were used to accelerate the convergence speed of the algorithm. Finally, a non-linear balance factor was used as the switching condition for the two stages, enabling more sufficient communication between populations and more effectively balancing global search and local exploitation. Xu Yifeng, Liu Sheng, Liu Yusong, et al. "Aquila Optimization Algorithm Integrated with Differential Mutation and Tangent Flight" [J]. Application Research of Computers, 2022, 39(10): 2996-3002. First, the mutation operation of differential evolution was introduced into the AO algorithm to make up for its deficiency in exploitation ability. Then, the tangent flight strategy with strong exploration ability in the tangent search algorithm was used to replace the Levy flight in the AO algorithm to increase the probability of the algorithm jumping out of the local optimum. Zhao J, Gao Z M, Chen H F. "The Simplified Aquila Optimization Algorithm" [J]. IEEE Access, 2022, 10: 22487-22515, simplified and adjusted the original Aquila optimization algorithm by abandoning the position update strategy in the exploitation process and only retaining the search stage.

[0005] To obtain an Aquila optimization algorithm with better global optimization performance and robustness, this application proposes a multi-strategy fusion improved Aquila optimization algorithm (Multi-Strategy Improved Aquila Optimizer, MSIAO) by improving the defects existing in its optimization method. Summary of the Invention

[0006] Aiming at the above-mentioned existing technologies, the technical problem to be solved by the present invention is the problem of uneven population initialization and being prone to falling into local optima in the traditional Aquila optimization algorithm for path planning.

[0007] To solve the above problems, the present invention provides a method for path planning of unmanned aerial vehicles with a hybrid multi-strategy, including the following steps: S1. Sort out the original Tianying optimization algorithm. The original Tianying optimization algorithm switches between these two stages by changing the number of iterations. When t ≤ 2T / 3, the algorithm executes the exploration stage; otherwise, it executes the exploitation stage; S2. Improve the Tianying optimization algorithm with multi-strategies; S3. Simulation results and analysis: Solve the benchmark test function, CEC2017 test function, and the path of the unmanned aerial vehicle in sequence to achieve the path planning of the unmanned aerial vehicle; S31. First, construct a mathematical model of the obstacle. By randomly generating the spatial coordinates of the obstacle, its specific position in the three-dimensional environment is determined; S32. Form a complete three-dimensional data description of the obstacle by assigning specific size information to the above coordinates; S33. Use a three-dimensional matrix to construct a three-dimensional model of the environment. In this model, the three-dimensional spatial range of the obstacle is assigned 1 to identify it as a non-flightable area; while the remaining non-obstacle areas are assigned 0 to represent the flightable space. This representation method not only simplifies the judgment of spatial relationships but also facilitates the processing of subsequent algorithms.

[0008] S34. Finally, use three-dimensional graphics rendering technology to map the interpolated terrain data into the three-dimensional space to form a three-dimensional terrain model with a sense of depth. Then, construct a mathematical model of the unmanned aerial vehicle's motion trajectory and plan a shortest path for the unmanned aerial vehicle from the starting point to the ending point; Specifically: The point sequence m = (m1, m2,..., m n ) = {(x0, y0, z0), (x1, y1, z1),..., (x n , y n , z n )} connecting the starting point and the ending point of the unmanned aerial vehicle is the moving path of the unmanned aerial vehicle, where the point sequence is composed of the optimal points of the decision-making target. Then, the calculation formula for the path length cost of the unmanned aerial vehicle is: (12).

[0009] As a further improvement of this application, the specific steps of step S1 are: S11. Population initialization: When establishing a mathematical model, the Tianying optimization algorithm first randomly initializes the positions of the population. The mathematical formula for this part is: (1) where rand is a random value from 0 to 1, UB i j represents the upper bound of the i-th Tianying in the j-th dimension, and LBi j Represents the lower bound of the i-th eagle in the j-th dimension, N represents the population size, and Dim represents the dimension of the search space; S11. Expanding exploration stage: In the first stage, the eagle flock expands the search range by soaring vertically at high altitude and hunts birds during flight. The mathematical formula for this part is: (2) Where X(t) and X(t + 1) represent the positions of individuals in the AO algorithm at the t-th and (t + 1)-th iterations, X best (t) represents the best individual in the 1 - t-th iteration, (1−t / T) controls the extended search, XM(t) represents the average position of the population at the t-th iteration, t represents the current iteration number, and T represents the maximum number of iterations; S13. Shrinking exploration stage: In the second stage, when the eagle flock discovers prey at high altitude, it forms a spiral hover above the prey. This is not only to more accurately locate the prey but also to prepare and build up momentum before landing. As the hover gradually tightens, an attack will be launched; its mathematical formula is: (3) Where X R (t) is a random individual within the population range [1, N], s is a constant, taking 0.01, u and v are random numbers in [0, 1], Levy(D) is the Levy flight distribution function, x, y represent the shape of the spiral flight, r is the search step size, taking a random integer within the range of 1 to 20, D1 is a vector from 1 to Dim, U takes 0.00565, ω takes 0.005, and θ1 = (3×π) / 2; S14. Expanding exploitation stage: In the third stage, when the eagle is about to launch an attack after hovering near the prey, it adopts a strategy of vertical descent. Its mathematical formula is: X3(t + 1)=(X best (t)−X M (t))×α−rand+((UB−LB)×rand+LB)×δ (4) Where α and δ represent the exploitation adjustment parameters, taking 0.1; S15. Shrinking exploitation stage: In this stage, the eagle approaches the prey: and launches a random attack on the prey. The mathematical formula for walking and grasping the prey is: (5) Among them, QF(t) is the quality function value used to balance the search strategy, G1 is various motion patterns when the eagle hunts prey, and G2 is the linearly decreasing flight slope value, with a range of [0, 2].

[0010] As a further improvement of this application, the specific steps of step S2 are as follows: S21. Set relevant parameters, the population size N, the maximum number of iterations T, the search dimension Dim, the search range UB, and UL; S22. Initialize the population using the Bernoulli chaos mapping strategy, S23. Calculate the individual fitness and select the currently optimal individual in terms of fitness; S24. Enter the population update stage and select formulas X1, X2, X3 * , X4 * for iterative optimization; S25. If the termination condition is met, stop the algorithm and output the optimization result; otherwise, return to step S3 to continue executing the algorithm.

[0011] As a further improvement of this application, the mathematical model of the Bernoulli chaos mapping in step S22 is as follows: (6) Among them, the chaos value Cn ∈ (0, 1), the initial value C1 = 0.7, and the control coefficient λ is 0.4; The value range of the global solution is [LB i , UB i , the i-th random number Ci ∈ [0, 1] generated by the Bernoulli chaos mapping, and the i-th initial position of the eagle generated by mapping to formula (1) using formula (7) is: X i * = C i × (UB - LB) + LB (7); In step S24, the shrinking encircling, spiral swimming, and predation mechanisms in the Whale Optimization Algorithm (WOA) are introduced to narrow the search range and improve the accuracy, and to develop the search space, enhance the coverage and blind spot cleaning, and improve the population diversity to improve the X3 model. Its mathematical formula is: (8) Among them, b is the logarithmic spiral shape constant, L is a random number between [-1, 1], and r1, r2 are random vectors in [0, 1]. In this process, the whale decides whether to swim towards the prey along a spiral curve or shrink the encircling circle towards the prey with a probability p between (0, 1); In step S24, the theft strategy of the Dung Beetle Optimizer (DBO) is integrated. By combining the influence of the global and local best positions and randomness, the global search and convergence speed of the algorithm are improved. At the same time, the global optimal guidance of the Tianying optimization algorithm is retained to promote the non-direct information exchange between solution sets, reduce the impact of randomness on robustness, and improve the late convergence speed and accuracy to improve the X4 model. Its mathematical formula is: (9) where g is a random vector subject to a normal distribution, S represents a constant set to 0.5, X * best represents the optimal position of the current iteration population, and Xbest represents the global optimal position.

[0012] As a further improvement of this application, in step S24, an efficient meta-heuristic algorithm needs to balance exploration and exploitation to avoid local optima. The original Tianying optimization algorithm switches phases by a fixed number of iterations, lacking information exchange between phases and is prone to falling into local optima. In this paper, a hybrid mutation strategy of Laplace and cosine transforms that dynamically adjusts with iterations is introduced to dynamically balance search and exploitation, determine premature convergence and jump out of local extrema, and improve the performance of the algorithm. Its mathematical formula is: (10) where η and µ take values of 3 and 0.08 respectively, and in the Laplace distribution Lap(a, b), the parameters a = 1 and b = 2; the original switching condition t ≤ 2T / 3 is replaced by rand ≤ λ to adaptively balance the global search and local exploitation phases of the algorithm; the value of λ decreases non-linearly with iterations, favoring global search and efficient exploration in the early stage, and local exploitation with a small probability to promote information exchange; in the later stage, it focuses on in-depth exploitation and occasional global search to prevent local optima. This non-linear balancing strategy effectively balances search and exploitation and improves the optimization speed and accuracy of the algorithm.

[0013] As another improvement of this application, the propeller blades of the unmanned aerial vehicle adopt variable pitch propeller blades. The variable pitch propeller blades include an H-shaped bracket, a propeller shaft fixedly connected to the middle of the H-shaped bracket, and two propeller blade bodies respectively connected to the clamping arms at both ends of the H-shaped bracket through magnetic control components. The propeller shaft is installed on the unmanned aerial vehicle through a rotating component.

[0014] As a supplement to another improvement of this application, the magnetic control component includes a regulation layer embedded with an angle sensor and two follower layers respectively fixedly connected to the upper and lower ends of the regulation layer. The two follower layers are respectively fixedly connected to the corresponding clamping arms of the H-shaped bracket, the propeller blade body is fixedly connected to the regulation layer, and a variable pitch unit is also arranged between the magnetic control component and the H-shaped bracket.

[0015] As another supplementary improvement of the present application, the pitch-changing unit includes two hemispherical columns respectively fixedly connected to the clamping arms of the H-shaped bracket by bolts, two groups of elastic columns respectively fixedly embedded in the follower layer, and two pairs of electromagnetic coils respectively fixedly installed in the upper and lower clamping arms of the H-shaped bracket. The electromagnetic coils are electrically connected to the power supply of the drone. The regulation layer is made of a magnetic material, and after the four electromagnetic coils are energized, a magnetic repulsive force is generated on the regulation layer. The follower layer is made of an elastic material.

[0016] In summary, by combining various strategies such as the Bernoulli chaos mapping strategy, the spiral stepping strategy, the stealing strategy integrating the dung beetle optimization algorithm, and the non-linear balance factor strategy to improve AO, a new MSIAO is obtained. Finally, through the non-linear balance factor strategy, the search and development are dynamically balanced, and the optimization speed and accuracy of the algorithm are improved. The MSIAO proposed by the present invention reduces the probability of the algorithm falling into local extrema, has higher optimization accuracy and faster convergence rate. In solving the UAV path planning problem, it can always find the shortest path compared with traditional methods. The present invention not only provides a new solution for UAV path planning, but also provides new ideas and methods for solving complex optimization problems.

[0017] And a variable-pitch propeller with a low mechanical structure is used to meet the different thrust requirements of the UAV in different situations during flight, effectively avoiding problems such as reduced accuracy and stuck pitch control caused by multiple mechanical structures. Combining the improvement of the algorithm in path planning and the improvement of the mechanical structure, the speed and stability of the UAV reaching the end point are greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the population initialization distribution diagram of the first embodiment of the present application; Figure 2 It is the flow chart of the improved eagle algorithm of the first embodiment of the present application; Figure 3 It is the result schematic diagram of the MSIAO of the first embodiment of the present application when solving a unimodal function; Figure 4 It is the comparison diagram of the path planning of each algorithm of the first embodiment of the present application; Figure 5 It is the benchmark test function data diagram of the first embodiment of the present application; Figure 6 It is the comparison diagram of the optimization results of different swarm intelligence algorithms of the first embodiment of the present application; Figure 7 It is the data diagram of the CEC2017 test function of the first embodiment of the present application; Figure 8 It is the optimization result diagram of the CEC2017 test function of the first embodiment of the present application; Figure 9 Ranking diagram of the MAE algorithm for the first implementation mode of this application; Figure 10 Data diagram of the path planning results of each algorithm for the first implementation mode of this application; Figure 11 Stereogram of the variable pitch blade for the second implementation mode of this application; Figure 12 Partial stereogram of the variable pitch blade for the second implementation mode of this application; Figure 13 Cross-sectional view of the front part of the magnetic control component for the second implementation mode of this application; Figure 14 Cross-sectional view of the side part of the magnetic control component for the second implementation mode of this application; Figure 15 Schematic diagram when the variable pitch blade of the unmanned aerial vehicle adjusts the blade in the second implementation mode of this application; Figure 16 Top view of the follower layer part in the second implementation mode of this application.

[0019] Explanation of the reference numerals in the figure: 1 Blade body, 2 Blade shaft, 3 H-shaped bracket, 4 Magnetic control component, 41 Regulation layer, 42 Follower layer, 51 Hemispherical column, 52 Elastic column, 6 Electromagnetic coil. Specific implementation mode

[0020] The following will make a detailed description of the two implementation modes of this application in conjunction with the accompanying drawings.

[0021] The first implementation mode: Figure 1 A method for path planning of an unmanned aerial vehicle using a hybrid multi-strategy is shown, including the following steps: S1. Sort out the original Tianying optimization algorithm. The original Tianying optimization algorithm switches between these two stages by changing the number of iterations. When t ≤ 2T / 3, the algorithm executes the exploration stage; otherwise, it executes the exploitation stage; The specific steps of step S1 are as follows: S11. Population initialization: When establishing a mathematical model for the Tianying optimization algorithm, first randomly initialize the positions of the population; the mathematical formula for this part is: (1) where rand is a random value from 0 to 1, UB i j represents the upper bound of the i-th Tianying in the j-th dimension, LB i j represents the lower bound of the i-th Tianying in the j-th dimension, N represents the population size, and Dim represents the dimension of the search space; S11. Expanding exploration stage: In the first stage, the eagle group expands the search range by soaring vertically at high altitude and hunts birds during flight. The mathematical formula for this part is: (2) where X(t) and X(t + 1) represent the positions of individuals in the AO algorithm at the t-th and (t + 1)-th iterations, X best (t) represents the best individual in the 1 - t-th iteration, (1−t / T) controls the extended search, XM(t) represents the average position of the population at the t-th iteration, t represents the current iteration number, and T represents the maximum number of iterations; S13. Shrinking exploration stage: In the second stage, when the eagle group discovers prey at high altitude, it forms a spiral hover above the prey. This is not only for more precise positioning of the prey but also for preparation and buildup before landing. As the hover gradually becomes more compact, an attack will be launched. The mathematical formula is: (3) where X R (t) is a random individual within the population range [1, N], s is a constant, taking 0.01, u and v are random numbers in [0, 1], Levy(D) is the Levy flight distribution function, x, y represent the shape of the spiral flight, r is the search step size, taking a random integer within the range of 1 to 20, D1 is a vector from 1 to Dim, U takes 0.00565, ω takes 0.005, and θ1 = (3×π) / 2; S14. Expanding exploitation stage: In the third stage, when the eagle is about to launch an attack after completing the hover near the prey, it adopts a strategy of vertical descent. The mathematical formula is: X3(t + 1)=(X best (t)−X M (t))×α−rand+((UB−LB)×rand+LB)×δ(4) where α and δ represent the exploitation adjustment parameters, taking 0.1; S15. Shrinking exploitation stage: In this stage, the eagle approaches the prey and launches a random attack on the prey. The mathematical formula for walking and grasping the prey is: (5) where QF(t) is the quality function value used to balance the search strategy, G1 is various motion patterns when the eagle tracks the prey, and G2 is the linearly decreasing flight slope value, with a range of [0, 2].

[0022] S2. Multi - strategy improved eagle optimization algorithm; The specific steps of step S2 are as follows: S21. Set relevant parameters, including population size N, maximum number of iterations T, search dimension Dim, search range UB, and UL; S22. Initialize the population using the Bernoulli chaotic mapping strategy; S23. Calculate the individual fitness and select the individual with the optimal current fitness; S24. Enter the population update stage and select formulas X1, X2, X3 * , X4 * for iterative optimization; S25. If the termination condition is met, stop the algorithm and output the optimization result; otherwise, return to step S3 to continue executing the algorithm.

[0023] As a further improvement of this application, the mathematical model of the Bernoulli chaotic mapping in step S22 is as follows: (6) where the chaotic value Cn ∈ (0, 1), the initial value C1 = 0.7, and the control coefficient λ is 0.4; The value range of the global solution is [LB i , UB i , the i-th random number Ci generated by the Bernoulli chaotic mapping ∈ [0, 1], and the i-th initial position of the eagle generated by mapping to formula (1) using formula (6) is: X i * = C i × (UB - LB) + LB (7); In step S24, the shrinking encircling, spiral swimming, and predation mechanisms in the Whale Optimization Algorithm (WOA) are introduced to narrow the search range, improve the accuracy, develop the search space, enhance the coverage and blind spot cleaning, and improve the population diversity to improve the X3 model. Its mathematical formula is: (8) where b is the logarithmic spiral shape constant, L is a random number between [-1, 1], and r1, r2 are random vectors between [0, 1]. In this process, the whale decides whether to swim towards the prey along a spiral curve or shrink the encircling circle towards the prey with a probability p between (0, 1); In step S24, the theft strategy of the Dung Beetle Optimizer (DBO) is integrated. By combining the influence and randomness of the global and local best positions, the global search and convergence speed of the algorithm are improved. At the same time, the global optimal guidance of the Tianying optimization algorithm is retained to promote the non-direct information exchange between solution sets, reduce the impact of randomness on robustness, and improve the later convergence speed and accuracy to improve the X4 model. Its mathematical formula is: (9) where g is a random vector subject to a normal distribution, S represents a constant set to 0.5, X * best represents the optimal position of the population in this iteration, and Xbest represents the global optimal position.

[0024] As a further improvement of this application, in step S24, an efficient meta-heuristic algorithm needs to balance exploration and exploitation to avoid local optima. The original Tianying optimization algorithm switches phases by a fixed number of iterations, lacking information exchange between phases and is prone to falling into local optima. In this paper, a hybrid mutation strategy of Laplace and cosine transform that dynamically adjusts with iterations is introduced to dynamically balance search and exploitation, determine premature convergence and jump out of local extrema, and improve the performance of the algorithm. Its mathematical formula is: (10) where η and μ take values of 3 and 0.08 respectively, and in the Laplace distribution Lap(a, b), the parameters a = 1 and b = 2; the original switching condition t ≤ 2T / 3 is replaced by rand ≤ λ to adaptively balance the global search and local exploitation phases of the algorithm; the value of λ decreases non-linearly with iterations, favoring global search and efficient exploration in the early stage, and occasional local exploitation with a small probability to promote information exchange; in the later stage, it focuses on in-depth exploitation and occasional global search to prevent local optima. This non-linear balancing strategy effectively balances search and exploitation and improves the optimization speed and accuracy of the algorithm.

[0025] S3. Simulation results and analysis: First: Solve the benchmark test function: To verify the feasibility of the newly proposed MSIAO algorithm, this test compares it with seven optimization algorithms, namely, the Artificial Ecosystem Optimization algorithm (AEO), Harris Hawk Optimization algorithm (HHO), Whale Optimization Algorithm (WOA), Particle Swarm Optimization algorithm (PSO), Butterfly Optimization Algorithm (BOA), Cuckoo Search Optimization Algorithm (COA), and Eagle Optimization Algorithm (AO). The population size and the number of iterations of all algorithms are set to 30 and 500 respectively, and the internal parameters follow the original settings. Ten benchmark functions are used, where f1 - f5 are unimodal functions to evaluate the convergence speed and accuracy, and f6 - f10 are multimodal functions to evaluate the global search ability and the ability to avoid local optima. The algorithms are run independently 100 times, and the performance of the algorithms is evaluated using the mean and standard deviation. A low mean indicates high convergence accuracy and strong optimization ability, while a low standard deviation reflects good algorithm stability. For details of the benchmark functions, see Figure 5 。

[0026] The MSIAO algorithm performs excellently on 10 benchmark test functions. From the Figure 6 optimization results, it shows significant advantages in both unimodal and multimodal functions. Compared with other algorithms, MSIAO has the highest convergence accuracy and strong stability in unimodal functions, approaching the theoretical optimal value. In multimodal complex functions such as F7, it also maintains high convergence accuracy and low standard deviation. Its average and standard deviation performance far exceeds other algorithms, proving that the improvement strategy significantly enhances the convergence accuracy and robustness of the algorithm, providing an efficient solution for complex optimization problems.

[0027] From Figure 3 it can be seen that when solving unimodal functions F1 - F4, MSIAO avoids the premature convergence problem of algorithms such as AEO, showing fast and stable convergence characteristics, approaching the theoretical optimal value. For F5, MSIAO is slightly inferior to HHO in the initial stage, but surpasses it in the middle and later stages, maintaining the highest accuracy. In multimodal functions, MSIAO is comparable to AO in the initial stage of F6, but MSIAO continues to progress without stagnation in the later stage and maintains the highest accuracy on F7. In F8 and F9, MSIAO leads with the fastest speed and the highest accuracy. On F10, MSIAO shows strong robustness by quickly converging and jumping out of local optima at multiple inflection points.

[0028] Generally speaking, MSIAO balances exploration and exploitation in global optimization, effectively avoiding local optima and showing excellent performance.

[0029] Second: Solving CEC2017 test functions: Refer to Figure 7 and Figure 8, ten CEC2017 benchmark functions were selected to verify the performance of MIAO. The results show that MIAO has the best average solution accuracy on 8 functions, especially excellent performance on hybrid functions, with significant convergence accuracy and robustness. Although it is not the best on 2 functions, the accuracy still ranks second. On the multimodal function CEC11, MIAO effectively avoids local optima, proving its powerful ability. Although its stability on unimodal functions is slightly inferior to that of COA, its accuracy is leading. Generally speaking, MIAO has both high accuracy and stability on most functions, successfully overcoming the deficiencies of the AO algorithm and improving the search efficiency and effect.

[0030] To comprehensively and concisely evaluate the competitiveness of MIAO compared with other algorithms, the mean absolute error MAE

[21] was used to rank the above algorithms. Figure 9 The MAE values and rankings of all algorithms for 10 CEC2017 benchmark test functions are shown, and its mathematical formula is: (11) where, NF is the number of functions; f i is the average value of the solution results on the i-th function; f i ' is the theoretical optimal value of the i-th function.

[0031] From Figure 9 it can be seen that the MIAO algorithm ranks first in terms of MAE, and the value is better than the other algorithms, comprehensively proving the effectiveness and stability of this algorithm.

[0032] Third: Solve the path of the UAV and realize the path planning of the UAV; S31. First, construct a mathematical model of the obstacle. By randomly generating the spatial coordinates of the obstacle, its specific position in the three-dimensional environment is determined; S32. Form a complete three-dimensional data description of the obstacle by assigning specific size information to the above coordinates; S33. Use a three-dimensional matrix to construct a three-dimensional model of the environment. In this model, the three-dimensional spatial range of the obstacle is assigned 1 to indicate the non-flightable area; while the remaining non-obstacle areas are assigned 0 to represent the flyable space; this representation method not only simplifies the judgment of spatial relationships but also facilitates the processing of subsequent algorithms.

[0033] S34. Finally, use three-dimensional graphics rendering technology to map the interpolated terrain data into the three-dimensional space to form a three-dimensional terrain model with a sense of depth. Then, construct a mathematical model of the UAV's motion trajectory and plan a shortest path for the UAV from the starting point to the ending point; Specifically: The UAV is connected by a sequence of points from the starting point to the ending point m=(m1,m2,...,m n) = {(x0, y0, z0), (x1, y1, z1),..., (x n , y n , z n )} is the path of the moving drone, where the point sequence consists of the optimal points of the decision-making goal; then the calculation formula for the path length cost of the drone is: (12).

[0034] To verify the performance of the improved algorithm MSIAO proposed in this paper in the path planning of drones, the following 7 algorithms are selected as comparison algorithms, and 10 path planning experiments are carried out with MSIAO in the same grid map environment. The uniformly specified experimental parameters: the population size and the number of iterations are 50 and 100 respectively, and the simulation experiment is carried out in a map of 1000×1000×120.

[0035] The performance indicators of the paths planned by each algorithm are compared in Figure 10 . It can be seen from Figure 10 that in the grid map, the average value of the optimal path of MSIAO is 1325.9633, which is the shortest among the 8 algorithms. The path length of MSIAO is shortened by 2.57% compared with WOA, 5.35% compared with COA, 7.79% compared with BOA, 10.95% compared with AO, 11.37% compared with HHO, 11.72% compared with AEO, and 17.44% compared with PSO. At the same time, MSIAO is also superior to other algorithms in terms of the minimum value of the path, so it has a more secure and reliable planning advantage. From Figure 4 , Figure 10 it can be seen that compared with the other 7 comparison algorithms, the path length of MSIAO is shorter and the path-finding performance is better.

[0036] In summary, the above improved Tianying optimization algorithm (MSIAO) with multi-strategy fusion improves AO by combining various strategies such as Bernoulli chaos mapping strategy, spiral step strategy, stealing strategy integrating dung beetle optimization algorithm, and nonlinear balance factor strategy to obtain the new MSIAO. Finally, the nonlinear balance factor strategy is used to dynamically balance search and development, improving the optimization speed and accuracy of the algorithm. The MSIAO proposed in the present invention reduces the probability of the algorithm falling into local extrema, has higher optimization accuracy and faster convergence rate. In solving the drone path planning problem, it can always find the shortest path compared with traditional methods. The present invention not only provides a new solution for drone path planning, but also provides new ideas and methods for solving complex optimization problems.

[0037] The second implementation method: On the basis of the first embodiment, this embodiment changes the drone into a drone with variable pitch blades, and the rest is the same as the first embodiment.

[0038] As Figures 11 - 12 , the blades of the drone adopt variable pitch blades. The variable pitch blades include an H-shaped bracket 3, a propeller shaft 2 fixedly connected to the middle of the H-shaped bracket 3, and two blade bodies 1 respectively connected to the inner sides of the two clamping arms at both ends of the H-shaped bracket 3 through magnetic control components 4. The propeller shaft 2 is installed on the drone through a rotating component. The specific rotating component is a prior art and will not be elaborated here.

[0039] As Figures 13 - 14 , the magnetic control component 4 includes a regulation layer 41 embedded with an angle sensor and two follower layers 42 respectively fixedly connected to the upper and lower ends of the regulation layer 41. The two follower layers 42 are respectively fixedly connected to the corresponding clamping arms of the H-shaped bracket 3, and the blade body 1 is fixedly connected to the regulation layer 41. A variable pitch unit is also provided between the magnetic control component and the H-shaped bracket 3. The variable pitch unit includes two hemispherical columns 51 respectively fixedly connected to the clamping arms of the H-shaped bracket 3 through bolts, two groups of elastic columns 52 respectively fixedly embedded in the follower layers 42, and two pairs of electromagnetic coils 6 respectively fixedly installed in the upper and lower clamping arms of the H-shaped bracket 3. The electromagnetic coils 6 are electrically connected to the power supply of the drone. The regulation layer 41 is made of a magnetic material, and after the four electromagnetic coils 6 are energized, a magnetic repulsive force is generated on the regulation layer 41. The follower layer 42 is made of an elastic material.

[0040] As Figure 15 , during use, multiple electromagnetic coils 6 are synchronously energized and turned on, thereby generating a magnetic repulsive force on the magnetic control component 4 in both the upper and lower directions. Under the guidance of the angle monitoring data of the magnetic control component 4 by the angle sensor, the repulsive force of the electromagnetic coils 6 on the magnetic control component 4 is adjusted to maintain the stability of the magnetic control component 4. When it is necessary to adjust the pitch angle of the blade body 1, the control center of the drone can control the repulsive force of different electromagnetic coils 6 on the magnetic control component 4 according to actual needs, so as to realize the up and down offset of the regulation layer 41 with the hemispherical column 51 as the fulcrum, and realize the change of the pitch angle of the blade body 1 connected to the magnetic control component 4, so as to adapt to different flight tasks of the drone.

[0041] Among them, the two hemispherical columns 51 are symmetrical to each other, and the end faces close to each other are both hemispherical and in contact with the center points of the corresponding end faces of the follower layer 42. As Figure 16 , the two electromagnetic coils 6 respectively correspond to the two edge vertices on the side of the blade body 1 close to the regulation layer 41. After the electromagnetic coils 6 are energized, the regulation layer 41 can be effectively maintained to offset up and down within a controllable range, and it is not easy to swing radially, improving its stability.

[0042] It should be noted that, in order to further improve the stability of the position of the magnetic control member 4, spherical grooves can be formed in the upper and middle parts of the upper and lower sides of the regulation layer 41, so that the two hemispherical columns 51 are respectively matched with the two spherical grooves.

[0043] Each set of elastic columns 52 are distributed in an annular array and there are no less than four. The arrangement of multiple elastic columns 52 can effectively improve the stability of the follower layer 42 and the blade body 1 connected to the follower layer 42, so that during flight, the position deviation is not likely to occur.

[0044] In the prior art, the pitch-changing blade is mostly regulated through the cooperation of multiple linkage members, such as cylinders, connecting rods, regulation discs, etc. Since the multiple mechanical structures are linked, wear is likely to occur and dust and impurities are likely to get stuck. As a result, when regulating the pitch angle, deviation is likely to occur, affecting the regulation accuracy. The pitch-changing blade of the present invention controls the angle change of the blade body 1 by means of magnetic control. Compared with the arrangement of multiple mechanical linkage moving parts in the prior art, the regulation accuracy is effectively guaranteed, and the regulation accuracy of the pitch angle is not likely to be affected by wear and dust jamming, which is convenient for adapting to different thrust requirements under different flights, and makes the UAV have higher stability and faster speed when flying along the path planned by the first embodiment towards the target point.

[0045] Combined with the current actual requirements, the above-mentioned embodiments adopted in the present application do not limit the protection scope thereto. Within the scope of knowledge possessed by those skilled in the art, various changes made without departing from the concept of the present application still fall within the protection scope of the present invention.

Claims

1. A method for path planning of unmanned aerial vehicles with a hybrid multi-strategy, characterized in that: It includes the following steps: S1. Sort out the original Tianying optimization algorithm; S2. Improve the Tianying optimization algorithm with multiple strategies; S3. Simulate the results and conduct path planning: Solve the benchmark test function, CEC2017 test function, and the path of the UAV in sequence to achieve the path planning of the UAV; S31. First, construct a mathematical model of the obstacle. By randomly generating the spatial coordinates of the obstacle, determine its specific position in the three-dimensional environment; S32. Form a complete three-dimensional data description of the obstacle by assigning specific size information to the above coordinates; S33. Use a three-dimensional matrix to construct a three-dimensional model of the environment. In this model, the three-dimensional space range of the obstacle is assigned 1 to indicate the non-flight area; while the remaining non-obstacle areas are assigned 0 to represent the flightable space; S34. Finally, use three-dimensional graphics rendering technology to map the interpolated terrain data into the three-dimensional space to form a three-dimensional terrain model with a three-dimensional sense. Then, construct a mathematical model of the UAV's motion trajectory and plan a shortest path for the UAV from the starting point to the ending point; Specifically: The sequence of points m=(m1, m2,..., m n )={(x0, y0, z0), (x1, y1, z1),...,(x n , y n , z n )} connecting the starting point and the ending point of the UAV is the path of the moving UAV, where the sequence of points consists of the optimal points of the decision-making objective; then the calculation formula for the path length cost of the UAV is: （12）。 2. The method for path planning of an unmanned aerial vehicle with a hybrid multi-strategy according to claim 1, wherein: The specific steps of step S1 are as follows: S11. Population initialization: When the Tianying optimization algorithm establishes a mathematical model, first randomly initialize the positions of the population; the mathematical formula for this part is: （1） where rand is a random value from 0 to 1, and UB i j represents the upper bound of the i-th eagle in the j-th dimension, and LB i j represents the lower bound of the i-th eagle in the j-th dimension, N represents the population size, and Dim represents the dimension of the search space; S11. Expand the exploration stage: In the first stage, the Tianying group expands the search range by soaring vertically at a high altitude and hunts birds during flight. The mathematical formula for this part is: （2） Among them, X(t) and X(t + 1) represent the positions of individuals in the t-th and (t + 1)-th iterations of the AO algorithm, and X best (t) represents the best individual in the (1 - t)-th iteration, (1 - t / T) controls the extended search, XM(t) represents the average position of the population at the t-th iteration, t represents the current iteration number, and T represents the maximum number of iterations; S13. Narrow the exploration stage: In the second stage, when the Tianying group discovers prey at a high altitude, it forms a spiral hover above the prey. As the hover gradually tightens, it will launch an attack; its mathematical formula is: （3） Among them, X R (t) is a random individual within the population range [1, N], s is a constant, taking 0.01, u and v are random numbers in [0, 1], Levy(D) is the Levy flight distribution function, x and y represent the shape of the spiral flight, r is the search step size, taking a random integer within the range of 1 to 20, D1 is a vector from 1 to Dim, U takes 0.00565, ω takes 0.005, and θ1 = (3×π) / 2; S14. Expand the exploitation stage: In the third stage, when the Tianying is about to launch an attack after hovering near the prey, it adopts a strategy of vertical descent. The mathematical formula for this is: X3(t + 1)=(X best (t) - X M (t)) × α - rand + ((UB - LB) × rand + LB) × δ (4) Among them, α and δ represent the exploitation adjustment parameters, taking 0.1; S15. Narrow the exploitation stage: In this stage, the Tianying approaches the prey and launches a random attack on the prey. The mathematical formulas for walking and grasping the prey are: （5） Among them, QF(t) is the quality function value used to balance the search strategy, G1 is various motion patterns when the Tianying tracks the prey, and G2 is the linearly decreasing flight slope value, with a range of [0, 2].

3. A method for path planning of an unmanned aerial vehicle with a hybrid multi-strategy according to claim 1, characterized in that: The specific steps of step S2 are as follows: S21. Set relevant parameters, population size N, maximum number of iterations T, search dimension Dim, search range UB, UL; S22. Initialize the population using the Bernoulli chaotic mapping strategy, S23. Calculate the individual fitness and select the current individual with the best fitness; S24. Enter the population update stage, and select formulas X1, X2, X3 according to the algorithm process * , X4 * for iterative optimization; S25. If the termination condition is met, stop the algorithm and output the optimization result; otherwise, return to step S3 to continue executing the algorithm.

4. A method for path planning of an unmanned aerial vehicle with a hybrid multi-strategy, as claimed in claim 3, wherein: The mathematical model of the Bernoulli chaotic mapping in step S22 is as follows: （6） Among them, the chaotic value Cn ∈ (0, 1), the initial value C1 = 0.7, and the control coefficient λ is 0.4; The value range of the global solution is [LB i , UB i . The \(i\)-th random number \(C_i\) generated by the Bernoulli chaotic map belongs to \([0, 1]\), and the \(i\)-th initial position of the Tianying generated by mapping to Equation (1) using Equation (6) is as follows: X i * =C i ×(UB−LB)+LB (7); In step S24, the whale optimization algorithm is introduced to improve the X3 model. The mathematical formula for this is: （8） Among them, b is the logarithmic spiral shape constant, L is a random number between [-1, 1], and r1, r2 are random vectors in [0, 1]; during this process, the whale decides whether to swim towards the prey along a spiral curve or shrink the encirclement and swim towards the prey with a probability p between (0, 1). In the step S24, the theft strategy of the dung beetle optimization algorithm is integrated to improve the X4 model, and its mathematical formula is: （9） where, g is a random vector subject to a normal distribution, S represents a constant set to 0.5, X * best represents the optimal position of the current iteration population, X best represents the global optimal position.

5. A method for path planning of an unmanned aerial vehicle with a hybrid multi-strategy according to claim 4, characterized in that: In the step S24, a hybrid mutation strategy of Laplace and cosine transform that dynamically adjusts with iterations is introduced to dynamically balance search and exploration, determine premature convergence and jump out of local extrema, and improve the algorithm performance. Its mathematical formula is: （10） Among them, η and μ take values of 3 and 0.08 respectively, and in the Laplace distribution Lap(a, b), the parameters a = 1 and b = 2; the original switching condition t ≤ 2T / 3 is replaced with rand ≤ λ to adaptively balance the global search and local exploration stages of the algorithm.

6. A drone used in conjunction with the hybrid multi-strategy drone path planning method according to claim 1, characterized in that: The propeller blades of the drone adopt variable pitch blades. The variable pitch blades include an H-shaped bracket, a propeller shaft fixedly connected to the middle of the H-shaped bracket, and two propeller blade bodies respectively connected to the inner sides of the two clamping arms at both ends of the H-shaped bracket through magnetic control components. The propeller shaft is installed on the drone through a rotating component.

7. A drone used in conjunction with the method for path planning of a drone with a hybrid multi-strategy according to claim 6, characterized in that: The magnetic control component includes a regulation layer embedded with an angle sensor and two follower layers respectively fixedly connected to the upper and lower ends of the regulation layer. The two follower layers are respectively fixedly connected to the corresponding clamping arms of the H-shaped bracket. The propeller blade body is fixedly connected to the regulation layer. A variable pitch unit is also arranged between the magnetic control component and the H-shaped bracket.

8. The unmanned aerial vehicle used in conjunction with the method for path planning of an unmanned aerial vehicle with a hybrid multi-strategy according to claim 7, characterized in that: The variable pitch unit includes two hemispherical columns respectively fixedly connected to the clamping arms of the H-shaped bracket through bolts, two groups of elastic columns respectively fixedly embedded in the follower layers, and two pairs of electromagnetic coils respectively fixedly installed in the upper and lower clamping arms of the H-shaped bracket. The electromagnetic coils are electrically connected to the power supply of the drone. The regulation layer is made of a magnetic material, and when the four electromagnetic coils are energized, they all generate magnetic repulsion forces on the regulation layer. The follower layer is made of an elastic material.

Images