Unmanned aerial vehicle path planning method of improved particle swarm algorithm based on multi-strategy fusion

Abstract

The invention discloses an unmanned aerial vehicle path planning method of an improved particle swarm algorithm based on multi-strategy fusion. The unmanned aerial vehicle path planning method comprises the following steps: 1) constructing a three-dimensional environment model; the method comprises the following steps of (1) building an algorithm fitness function, (2) initializing a population position, (4) initializing a particle swarm algorithm parameter, (5) starting iteration, (6) updating the position, (7) dynamically and adaptively updating the parameter, (8) carrying out a random walk strategy and (9) judging. The method has very strong global planning and local planning capabilities, so that the unmanned aerial vehicle can adapt to various environments to work, and has very strong practicability.

Description

Technical Field

[0001] The invention relates to UAV technology, in particular to a UAV path planning method based on an improved particle swarm algorithm with multi-strategy fusion. Background Art

[0002] Due to the characteristics of high maneuverability and easy operation, UAVs have attracted much attention in many fields such as environmental monitoring and earthquake relief. Among them, UAV path planning technology is of great significance in improving the overall system performance of UAVs. By planning the path of UAVs, the work efficiency of UAVs can be significantly improved, while reducing energy consumption and costs. Among many path planning algorithms, the particle swarm algorithm has the characteristics of good robustness and fast convergence speed, but the particle swarm algorithm still has the disadvantages of low search efficiency and easy to fall into local optimality, and the local planning ability of the particle swarm algorithm is not strong. Balancing the global and local planning capabilities of the algorithm is the main task at present. The existing technologies are mostly linear improvements in learning factors, inertia weights, and improving population initialization positions. These methods are not the best for improving the particle swarm algorithm. Secondly, there is a lack of technology to comprehensively improve the particle swarm algorithm to improve the overall performance of the algorithm. Summary of the invention

[0003] The purpose of the present invention is to provide a UAV path planning method based on an improved particle swarm algorithm with multi-strategy fusion to address the deficiencies of the prior art. This method has strong global planning and local planning capabilities, enabling the UAV to adapt to a variety of environments for operation and has strong practicality.

[0004] The technical solution for achieving the purpose of the present invention is:

[0005] A UAV path planning method based on an improved particle swarm algorithm with multi-strategy fusion includes the following steps:

[0006] 1) Constructing a 3D environmental model: Using exponential functions to simulate the mountain environment during the field operation of UAVs, a 3D environmental model is constructed;

[0007] 2) Establishing the algorithm fitness function: The algorithm fitness function is established based on the flight path length, flight angle and energy consumption of the drone. The fitness function is composed of path length factor, flight angle factor and energy consumption factor. The formula is as follows:

[0008] J=ζ1×Length+ζ2×Angle+ζ3×Εnergy,

[0009] In the formula, ζ1, ζ2, ζ3 are weighted coefficients and satisfy ζ1+ζ2+ζ3=1; J is the total fitness function; Length is the path length function; Angle is the flight angle function; Energy is the energy consumption function, and the calculation process is:

[0010] 2-1) Calculate the path length function: The calculation method is:

[0011]

[0012] In the formula, S is the i-th node (x i ,y i ,z i ) and the i+1th node (x i+1 ,y i+1 ,z i+1 ) between the two sides;

[0013] 2-2) Calculate the flight angle function: The formula is as follows:

[0014]

[0015] ν1=(x i -x i-1 ,y i -y i-1 ,z i -z i-1 ),

[0016] ν2=(x i+1 -x i ,y i+1 -y i ,z i+1 -z i ),

[0017] In the formula, ν1 and ν2 are the node vectors between the ith node and the previous node and the next node;

[0018] 2-3) Calculate energy consumption: The formula is as follows:

[0019]

[0020] Energy=E1+E2,

[0021] In the formula, E1 is the energy consumed by the drone in the vertical direction; E2 is the energy consumed by the drone in the horizontal direction;

[0022] 3) Initialize the population position: Use the elite reverse learning strategy to initialize the population position. This strategy is used to construct an elite particle group, which increases the search range of particles, avoids the algorithm from falling into the local optimum, and improves the quality of the initial solution. Specifically, the elite reverse solution formula is used to obtain the elite individual of the tth iteration. The formula is as follows:

[0023]

[0024] Among them, t is the number of iterations, a and b are the upper and lower bounds of the environment model respectively;

[0025] 4) Initialize the parameters of the particle swarm algorithm: The parameters of the particle swarm algorithm include the initial velocity v of the particles, the inertia weight ω, the individual learning factor c1, the social learning factor c2, the population size N, and the maximum number of iterations T of the algorithm. max ;

[0026] 5) Start iteration: Calculate the path fitness value according to the fitness function established in step 2) to obtain the optimal value and optimal position of the population;

[0027] 6) Position update: The golden sine strategy is used to improve the particle position update formula to avoid blind global search of the algorithm, enhance the particle optimization speed and local search ability, and then the particle updates its position according to the optimal position of the population and the new formula. The specific method is as follows:

[0028] h1=-3·(1-τ)+3·τ,

[0029] h2=-3·τ+3·(1-τ),

[0030] τ=0.618,

[0031] P i t+1 =P i t ·|sin(θ1)|-θ2·sin(θ1)·|h1·g t best -h2·P i t |,

[0032] Among them, τ is the golden ratio, θ1 determines the moving distance of the individual in the next iteration, and its value range is [0,2π]; θ2 determines the position update direction of the individual in the next iteration, and its value range is [0,π]; g t best is the optimal position of the particle population at the tth iteration;

[0033] 7) Dynamic adaptive update parameters: According to the current number of iterations, the social learning factor, individual learning and factor inertia weight are dynamically and adaptively updated. The formula for updating the social learning factor and individual learning factor is as follows:

[0034]

[0035] Among them, c1 is the individual learning factor; c2 is the social learning factor; t max is the maximum number of iterations;

[0036] The formula for updating the inertia weight is as follows:

[0037]

[0038] Among them, ω is the inertia weight;

[0039] 8) Random walk strategy: After the particle position is updated, the random walk strategy is used to compare with the updated particle position. According to the greedy strategy, the position with better fitness is selected to enhance the ability of the algorithm to jump out of the local optimal solution. The random walk strategy is as follows:

[0040] L'(t)=L(t)+ε·step·η,

[0041] In the formula, step is the walking step length, ε is the scaling factor, the value range is [0,1], η is the random walking direction, the value range is [0,2π];

[0042] 9) Judgment: Determine whether the number of iterations has reached the maximum. If so, update and output the optimal solution and the algorithm ends. If not, go to step 6).

[0043] In this technical solution, the elite reverse learning strategy can improve the quality of the initial solution, expand the search range, and enable the algorithm to jump out of the local optimal solution. The golden sine strategy can balance the global planning and local planning capabilities of the algorithm. The random walk strategy and greedy strategy can prevent the algorithm from falling into the local optimal solution. The overall performance of the particle swarm algorithm after the fusion of multiple strategies will be greatly improved.

[0044] Compared with the standard particle swarm algorithm, the advantages of this technical solution are:

[0045] 1. When this technical solution is applied to UAV path planning, the initial solution is better, which greatly improves the search efficiency of the algorithm;

[0046] 2 The integration of multiple strategies enables this technical solution to have a strong ability to escape from local optimality, and it also balances the global planning and local planning capabilities of the algorithm, improves the overall performance of the algorithm, and makes the algorithm more practical.

[0047] This method has strong global planning and local planning capabilities, allowing the drone to adapt to a variety of environments and is very practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of a method flow chart of an embodiment;

[0049] Figure 2 An environmental model for an embodiment;

[0050] Figure 3 This is a schematic diagram showing the fitness comparison between the population initialized by the elite reverse learning strategy and the standard particle swarm algorithm in this embodiment. DETAILED DESCRIPTION

[0051] The content of the present invention is further described below in conjunction with the drawings and embodiments, but the present invention is not limited thereto.

[0052] Example:

[0053] Reference Figure 1 , a UAV path planning method based on an improved particle swarm algorithm with multi-strategy fusion, including the following steps:

[0054] 1) Construct a 3D environment model: Use an exponential function to simulate the mountain environment during the field operation of the UAV, and then construct a 3D environment model such as Figure 2 As shown;

[0055] 2) Establishing the algorithm fitness function: The algorithm fitness function is established based on the flight path length, flight turning angle and energy consumption of the drone. The fitness function is composed of path length factor, flight turning angle factor and flight altitude factor. The formula is as follows:

[0056] J=ζ1×Length+ζ2×Angle+ζ3×Εnergy,

[0057] In the formula, ζ1, ζ2, ζ3 are weighted coefficients and satisfy ζ1+ζ2+ζ3=1; J is the total fitness function; Length is the path length function; Angle is the flight angle function; Energy is the energy consumption function, and the calculation process is:

[0058] 2-1) Calculate the path length function: The calculation method is:

[0059]

[0060] In the formula, S is the i-th node (x i ,y i ,z i ) and the i+1th node (x i+1 ,y i+1 ,zi+1 ) between the two sides;

[0061] 2-2) Calculate the flight angle function: The formula is as follows:

[0062]

[0063] ν1=(x i -x i-1 ,y i -y i-1 ,z i -z i-1 ),

[0064] ν2=(x i+1 -x i ,y i+1 -y i ,z i+1 -z i ),

[0065] In the formula, ν1 and ν2 are the node vectors between the ith node and the previous node and the next node;

[0066] 2-3) Calculate energy consumption: The formula is as follows:

[0067]

[0068] Energy=E1+E2,

[0069] In the formula, E1 is the energy consumed by the drone in the vertical direction; E2 is the energy consumed by the drone in the horizontal direction;

[0070] 3) Initialize the population position: Use the elite reverse learning strategy to initialize the population position. This strategy is used to construct an elite particle group, which increases the search range of particles, avoids the algorithm from falling into the local optimum, and improves the quality of the initial solution. Specifically, the elite reverse solution formula is used to obtain the elite individual of the tth iteration. The formula is as follows:

[0071]

[0072] Among them, t is the number of iterations, a and b are the upper and lower bounds of the environment model respectively; Figure 3 As shown in the figure, the initial population quality of the algorithm is significantly improved after the elite reverse learning strategy is adopted in this example. The fitness value of the initial solution is about 176, while the fitness function value of the initial solution of the traditional particle swarm algorithm is about 230. The quality of the initial solution is greatly improved. Moreover, after the elite reverse learning strategy is adopted, the algorithm converges in about 40 times, while the traditional particle swarm algorithm needs about 52 times, which speeds up the convergence of the algorithm to a certain extent.

[0073] 4) Initialize the parameters of the particle swarm algorithm: The parameters of the particle swarm algorithm include the initial velocity of particles, inertia weight, social learning factor, individual learning factor, population size, and number of algorithm iterations;

[0074] 5) Start iteration: Calculate the path fitness value according to the fitness function established in step 2) to obtain the optimal value and optimal position of the population;

[0075] 6) Position update: The golden sine strategy is used to improve the particle position update formula to avoid blind global search of the algorithm, enhance the particle optimization speed and local search ability, and then the particle updates its position according to the optimal position of the population and the new formula. The specific method is as follows:

[0076] h1=-3·(1-τ)+3·τ,

[0077] h2=-3·τ+3·(1-τ),

[0078] τ=0.618,

[0079] P i t+1 =P i t ·|sin(θ1)|-θ2·sin(θ1)·|h1·g t best -h2·P i t |,

[0080] Among them, τ is the golden ratio, θ1 determines the moving distance of the individual in the next iteration, and its value range is [0,2π]; θ2 determines the position update direction of the individual in the next iteration, and its value range is [0,π]; g t best is the optimal position of the particle population at the tth iteration;

[0081] 7) Dynamic adaptive update parameters: According to the current number of iterations, the social learning factor, individual learning and factor inertia weight are dynamically and adaptively updated. The formula for updating the social learning factor and individual learning factor is as follows:

[0082]

[0083] Among them, c1 is the individual learning factor; c2 is the social learning factor; t max is the maximum number of iterations;

[0084] The formula for updating the inertia weight is as follows:

[0085]

[0086] Among them, ω is the inertia weight;

[0087] 8) Random walk strategy: After the particle position is updated, the random walk strategy is used to compare with the updated particle position. According to the greedy principle, the position with better fitness is selected to enhance the ability of the algorithm to jump out of the local optimal solution. The random walk strategy is as follows:

[0088] L'(t)=L(t)+ε·step·η, where step is the step length, ε is the scaling factor, and its value range is [0,1]; η is the random walk direction, and its value range is [0,2π];

[0089] 9) Judgment: Determine whether the number of iterations has reached the maximum. If so, update and output the optimal solution and the algorithm ends. If not, go to step 6).

Claims

1. A UAV path planning method based on an improved particle swarm algorithm with multi-strategy fusion, characterized in that: The steps include: 1) Constructing a 3D environmental model: Using exponential functions to simulate the mountain environment during the field operation of UAVs, a 3D environmental model is constructed; 2) Establishing the algorithm fitness function: The algorithm fitness function is established based on the flight path length, flight angle and energy consumption of the drone. The fitness function is composed of path length factor, flight angle factor and energy consumption factor. The formula is as follows: J=ζ1×Length+ζ2×Angle+ζ3×Εnergy, In the formula, ζ1, ζ2, and ζ3 are weighted coefficients and satisfy ζ1+ζ2+ζ3=1; J is the total fitness function; Length is the path length function; Angle is the flight angle function; Energy is the energy consumption function, and the calculation process is: 2-1) Calculate the path length function. The calculation method is: In the formula, S is the i-th node (x i ,y i ,z i ) and the i+1th node (x i+1 ,y i+1 ,z i+1 ) between the two sides; 2-2) Calculate the flight angle function, the formula is as follows: ν1=(x i -x i-1 ,y i -y i-1 ,z i -z i-1 ), ν2=(x i+1 -x i ,y i+1 -y i ,z i+1 -z i ), In the formula, ν1 and ν2 are the node vectors between the ith node and the previous node and the next node; 2-3) Calculate energy consumption: The formula is as follows: Energy=E1+E2, In the formula, E1 is the energy consumed by the drone in the vertical direction; E2 is the energy consumed by the drone in the horizontal direction; 3) Initialize the population position: Use the elite reverse learning strategy to initialize the population position. Specifically, use the elite reverse solution formula to obtain the elite individuals of the tth iteration, and then use this to construct a new population. The reverse solution formula is as follows: Among them, t is the number of iterations, a and b are the upper and lower bounds of the environment model respectively; 4) Initialize the parameters of the particle swarm algorithm: The parameters of the particle swarm algorithm include the initial velocity v of the particles, the inertia weight ω, the individual learning factor c1, the social learning factor c2, the population size N, and the maximum number of iterations T of the algorithm. max ; 5) Start iteration: Calculate the path fitness value according to the fitness function established in step 2) to obtain the optimal value and optimal position of the population; 6) Position update: The golden sine strategy is used to improve the particle position update formula. Then the particles update their positions according to the optimal position of the population and the new formula. The specific method is as follows: h1=-3·(1-τ)+3·τ, h2=-3·τ+3·(1-τ), τ=0.618, P i t+1 =P i t ·|sin(θ1)|-θ2·sin(θ1)·|h1·g t best -h2·P i t |, Among them, τ is the golden ratio, θ1 determines the moving distance of the individual in the next iteration, and its value range is [0,2π]; θ2 determines the position update direction of the individual in the next iteration, and its value range is [0,π]; g t best is the optimal position of the particle population at the tth iteration; 7) Dynamic adaptive update parameters: According to the current number of iterations, the social learning factor, individual learning and factor inertia weight are dynamically and adaptively updated. The formula for updating the social learning factor and individual learning factor is as follows: Among them, c1 is the individual learning factor; c2 is the social learning factor; t max is the maximum number of iterations; The formula for updating the inertia weight is as follows: Among them, ω is the inertia weight; 8) Random walk strategy: After the particle position is updated, the random walk strategy is used to compare with the updated particle position. According to the greedy principle, a position with better fitness is selected. The random walk strategy is as follows: L'(t)=L(t)+ε·step·η, In the formula, step is the walking step length, ε is the scaling factor, the value range is [0,1], η is the random walking direction, the value range is [0,2π]; 9) Judgment: Determine whether the number of iterations has reached the maximum. If so, update and output the optimal solution and the algorithm ends. If not, go to step 6).

Images