AGV path planning method based on improved HHO algorithm integrated with neural network

Abstract

The present invention discloses an AGV path planning method of an improved HHO algorithm fused with a neural network. The path smoothing process adopts a Bezier curve, which can eliminate sharp inflection points to make it more suitable for actual use of AGV. The environment modeling is modeled by a visual graph method, which uses convex polygons to fit the obstacle information of the mapping, thereby simplifying the path planning problem to the problem of finding the shortest route under constraints. The beneficial effects of the present invention are: the AGV path planning method of the improved HHO algorithm fused with a neural network, the improved HHO algorithm converges faster than the HHO algorithm, and obtains a shorter and smoother path.

Description

Technical Field

[0001] The present invention relates to the field of AGV path planning, and in particular to an AGV path planning method integrating an improved HHO algorithm with a neural network. Background Art

[0002] The A* algorithm is a concept proposed by foreign expert PE Hart. It is a more effective method for solving path search problems in static path planning and has been applied to fields such as intelligent guided vehicles and GPS navigation systems. The main idea of ​​this method is: taking the starting point of the AGV handling task as the center and all the surrounding walkable grids as the target point, it gradually expands to the peripheral area until the extended object is the target point and the algorithm is completed. The A* algorithm is widely used for its advantages such as wide search range and multi-index evaluation path. However, it is not effective in studying the path planning problem of workshop systems. The A* algorithm needs to decompose the environment into a large number of grid nodes as data sets for formula calculation in the process. This decomposition method is too complicated for the workshop AGV operation environment. This process will consume a lot of time in the algorithm operation and cannot fully guarantee the search for the optimal path. Once the workshop environment changes, all data sets will need to be redefined, which brings great trouble to the optimization problem. As for AGV operation, the AGV operation angle cannot be controlled, which will also cause difficulties in laying specific paths in actual production.

[0003] The intelligent algorithm that has attracted much attention from researchers at present is the Harris Hawk Optimization Algorithm (HHO), which has the characteristics of few parameters and high precision. Many scholars at home and abroad have used this algorithm to solve some practical problems. Essam et al. used the Harris Hawk algorithm for drug optimization design. Aneesh et al. introduced an adaptive mechanism based on HHO to study the image threshold segmentation problem. The traditional Harris Hawk algorithm is not perfect, and the structure itself has certain defects, that is, the problem of convergence to local extreme points that most bio-inspired algorithms have. Summary of the invention

[0004] The purpose of the present invention is to provide an AGV path planning method of an improved HHO algorithm integrating a neural network to solve the problems raised in the above background technology.

[0005] The purpose of the present invention is to achieve this through the following technical solution: an AGV path planning method of an improved HHO algorithm integrating a neural network, comprising the following steps:

[0006] (1) Step 1: Initialization phase. In the initialization phase, an obstacle map is generated, and the position of the Harris hawk individual is initialized after the algorithm-related parameters are set, and a preliminary path is obtained;

[0007] (2) The second step: iteration phase. In the iteration phase, the path is first smoothed, the escape energy and the fitness value of the path are calculated, and the path is subjected to collision detection against all obstacles to determine whether it collides with the obstacle.

[0008] (3) The third step: judgment phase. The judgment phase mainly updates the individual position and escape energy according to the iteration results until the maximum number of times is reached and the optimal path is output;

[0009] In a preferred embodiment of the present invention, the path smoothing process uses a Bezier curve, which can eliminate sharp inflection points, making it more suitable for actual use of AGV.

[0010] In a preferred embodiment of the present invention, the environment modeling is performed by a visual graph method, which uses convex polygons to fit the mapped obstacle information, thereby simplifying the path planning problem into a problem of finding the shortest route under constraints.

[0011] In a preferred embodiment of the present invention, in the HHO algorithm, the size of the prey energy E plays an important role in regulating and transitioning random detection and local siege, when |E|<1 and the smaller it is, the algorithm performs local siege; when |E|≥1 and the larger it is, the algorithm performs random detection; the energy equation E of the traditional HHO decreases linearly from the initial value E0 to 0, and this change process is not conducive to convergence away from the local extreme value; therefore, the global and local searches are balanced with dynamic energy, and the idea of ​​the hyperbolic tangent function is first borrowed to introduce a dynamic hyperbolic tangent rate S at the position update in the early and middle stages of the algorithm iteration, and a sinusoidal fluctuation rate is introduced in the late stage of the iteration; the dynamic hyperbolic tangent rate used, in the early stage of the algorithm iteration, the improved algorithm obtains a larger prey energy to ensure that it searches for the optimal solution in a wider area; in the middle stage of the algorithm iteration, the improved algorithm obtains a smaller prey energy, so that the current individual can perform accurate search in the optimal individual, effectively balancing global detection and local optimization; in the late stage of the algorithm iteration, the characteristics of the irregular transformation of the sine wave are used to enhance the diversity of the optimal individual in the local space development, and assist the Harris hawk individual to jump out of the local optimum.

[0012] The beneficial effects of the present invention are as follows: the AGV path planning method of the improved HHO algorithm of the fusion neural network has a faster convergence speed than the HHO algorithm, and a shorter and smoother path is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a system diagram of the present invention;

[0014] Figure 2 It is a dynamic energy change curve diagram of the present invention;

[0015] Figure 3 It is the environmental model diagram of the present invention;

[0016] Figure 4 It is a comparison diagram of the static environment planning path of the present invention;

[0017] Figure 5 It is the dynamic environment planning path diagram of the present invention. DETAILED DESCRIPTION

[0018] The following is a further description in conjunction with specific embodiments, but the specific embodiments below should not be understood as limiting the main invention. Various changes and modifications that can be obviously made by ordinary technicians in this field based on the present invention should all be within the scope of the invention.

[0019] like Figure 1 , embodiments of the present invention include:

[0020] The AGV path planning method based on the improved HHO algorithm integrating neural network consists of the following steps:

[0021] (1) Step 1: Initialization phase. In the initialization phase, an obstacle map is generated, and the position of the Harris hawk individual is initialized after the algorithm-related parameters are set, and a preliminary path is obtained;

[0022] (2) The second step: iteration phase. In the iteration phase, the path is first smoothed, the escape energy and the fitness value of the path are calculated, and the path is subjected to collision detection against all obstacles to determine whether it collides with the obstacle.

[0023] (3) The third step: judgment phase. The judgment phase mainly updates the individual position and escape energy according to the iteration results until the maximum number of times is reached and the optimal path is output;

[0024] It is worth mentioning that the path smoothing process uses Bezier curves, which can eliminate sharp turning points and make it more suitable for the actual use of AGV. The environment modeling is modeled by visual graph method, which uses convex polygons to fit the mapped obstacle information, thereby simplifying the path planning problem to the problem of finding the shortest route under constraints. In the HHO algorithm, the size of the prey energy E plays an important role in regulating and transitioning between random detection and local siege. When |E|<1 and the smaller it is, the algorithm performs local siege; when |E|≥1 and the larger it is, the algorithm performs random detection; the energy equation E of the traditional HHO decreases linearly from the initial value E0 to 0, and this change process is not conducive to escaping from the local extreme value. convergence; therefore, the global and local searches are balanced with dynamic energy. Firstly, the idea of ​​hyperbolic tangent function is borrowed. The dynamic hyperbolic tangent rate s is introduced in the position update in the early and middle stages of the algorithm iteration, and the sine wave rate is introduced in the later stages of the iteration; the dynamic hyperbolic tangent rate used in it, in the early stage of the algorithm iteration, the improved algorithm obtains a larger prey energy to ensure that it searches for the optimal solution in a wider area; in the middle stage of the algorithm iteration, the improved algorithm obtains a smaller prey energy, so that the current individual can perform precise search in the optimal individual, effectively balancing the global detection and local optimization; in the later stage of the algorithm iteration, the characteristics of irregular transformation of the sine wave are used to enhance the diversity of the optimal individual in the local space development, and assist the Harris hawk individual to jump out of the local optimum.

[0025] Specific implementation method: improved Harris Hawk optimization algorithm, (1) dynamic energy strategy, in the HHO algorithm, the size of prey energy E plays an important role in regulating and transitioning random detection and local siege, when |E|<1 and the smaller it is, the algorithm performs local siege; when |E|≥1 and the larger it is, the algorithm performs random detection; the energy equation E of the traditional HHO decreases linearly from the initial value E0 to 0, and this change process is not conducive to convergence away from the local extreme value; therefore, the global and local searches are balanced by dynamic energy, firstly, the idea of ​​the hyperbolic tangent function is borrowed, and the dynamic hyperbolic tangent rate S is introduced at the position update in the early and middle stages of the algorithm iteration, and the sine wave rate is introduced in the late stage of the iteration; the dynamic hyperbolic tangent rate used, in the early stage of the algorithm iteration, the improved algorithm obtains a larger prey energy to ensure that it searches for the optimal solution in a wider area; in the middle stage of the algorithm iteration, the improved algorithm obtains a smaller prey energy, so that the current individual can perform accurate search in the optimal individual, effectively balancing global detection and local optimization; in the late stage of the algorithm iteration, the characteristics of the irregular transformation of the sine wave are used to enhance the diversity of the optimal individual in the local space development, and assist the Harris Hawk individual to jump out of the local optimum;

[0026]

[0027] Where S represents the dynamic hyperbolic tangent rate; S 0 represents the initial velocity, t maxrepresents the maximum number of iterations, t represents the number of iterations; S end Indicates the speed at the end of the iteration; when the number of iterations T = 0, S 0 =3, S end =0.4 indicates the speed at the end of the iteration, that is, the number of iterations reaches the maximum t max ; Use adjustment factors to control the smoothness of the curve, using μ 1 =0.23, μ 2 =0.06, μ 3 =1.6, θ=0.3, the simulation results are optimal;

[0028] (2) Curve smoothing and calculation. Bezier curve is a commonly used vector drawing curve that can draw any number of points in a two-dimensional plane into a smooth curve. The introduction of Bezier curve can eliminate sharp inflection points, making it more suitable for actual use of AGV.

[0029] The expression of the n-order Bezier curve is as follows:

[0030]

[0031] In the formula, P(s) is the control point, s is the variable, P(i) represents the position point, and B i,n (s) is the n-th degree Bernstein polynomial, which can be expressed as:

[0032]

[0033] In the formula, the number of position points is n+1, and the number of control points is n-1;

[0034] The first-order derivative formula of the Bezier curve is as follows:

[0035]

[0036] The length of the smoothed curve also needs to calculate the fitness value. The smoothed path is divided into H equal parts, with the path length F i Is the fitness value, then the fitness function is as follows:

[0037]

[0038] Where ||h i,k+1 -h i,k || is the end point h i,k+1 and starting point h i,k The distance between

[0039] (3) Simulation and analysis, environmental modeling. The present invention uses visual graph method to model the obstacle information using convex polygon fitting mapping, thereby simplifying the path planning problem into the problem of finding the shortest route under constraints; Figure 3It is an environmental model for AGV path planning. In the model, obstacles are mapped into polygons of various shapes. The ray connecting the starting point S to the end point T is the horizontal axis, and the vertical line through the starting point is the vertical axis. A local coordinate system S-X'Y' is established. D virtual coordinates are used to divide the path into D+1 segments with vertical lines L1~LD on the horizontal axis. i (i=1,2,…,D) randomly select a collision-free node P i , and finally connect the points to get the complete path P patp = {P 1 -S,…,P D -P D-1 ,TP D}; Then use the formula to transform the collision-free nodes into the global coordinate system coordinates. The conversion equation is as follows:

[0040]

[0041] Where (x s ,y x ) is the coordinate of the starting point S in the global coordinate system, and the x coordinate of each path point can be obtained by the formula Obtained in the local coordinate system; θ is the angle at which the global coordinate system x-axis is rotated counterclockwise to be parallel to the local coordinate system x';

[0042] (4) Parameter setting. To verify the superiority of the improved algorithm in the AGV path planning problem, the algorithm idea was simulated based on Matlab R2016a software. The algorithm parameters were set as follows: population size N = 100; maximum number of iterations tmax = 200; initial escape energy of prey E0 = 3, and the simulation environment range was 100 m × 100 m.

[0043] (5) Static map simulation: setting up two different static environments with fewer obstacles and more obstacles for simulation comparison; Figure 4 A comparison of the path planning between the traditional HHO algorithm and the improved HHO algorithm under environments with fewer obstacles and more obstacles is given;

[0044] (6) Dynamic map simulation: Dynamic obstacles are set at the map coordinates (95, 20) and (60, 95) respectively. The circular obstacle moves along the negative direction of the x-axis at the same speed as the AGV until it touches the y-axis, and the rectangular obstacle moves along the negative direction of the y-axis at the same speed as the AGV until it touches the x-axis. Repeat 30 times to obtain the optimal result path, such as Figure 5 As shown;

[0045] (7) Result analysis, the simulation results are summarized in Table 1;

[0046] Table 1 Index comparison

[0047]

[0048] The simulation results show that in a static environment with few obstacles, the optimal path of the improved HHO algorithm is 18.41m shorter than that of the traditional HHO algorithm, which is 11.87% shorter, and the optimal number of iterations is reduced by 26 times, about 14.61%; in a static environment with many obstacles, the optimal path of the improved HHO algorithm is 58.97m shorter than that of the traditional HHO algorithm, which is 33.86% shorter, and the optimal number of iterations is reduced by 30 times, about 16.22%; in a dynamic obstacle environment, the improved HHO plans an optimal path of 115.26 and a stable number of iterations of 165. Since the traditional HHO algorithm does not integrate neural networks and obstacle avoidance detection, the failure rate in the dynamic environment is extremely high, so it is not compared here; Overall, the improved HHO algorithm converges faster than the HHO algorithm and obtains a shorter and smoother path.

[0049] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. The AGV path planning method based on the improved HHO algorithm integrating neural network is characterized by: It consists of the following steps: (1) Step 1: Initialization phase. In the initialization phase, an obstacle map is generated, and after setting the algorithm-related parameters, the position of the Harris hawk individual is initialized and a preliminary path is obtained. (2) The second step: iteration phase. In the iteration phase, the path is first smoothed, the escape energy and the fitness value of the path are calculated, and the path is subjected to collision detection against all obstacles to determine whether it collides with the obstacle. (3) The third step: judgment phase. The judgment phase mainly updates the individual position and escape energy according to the iteration results until the maximum number of times is reached and the optimal path is output; In the HHO algorithm, the size of prey energy E plays an important role in regulating and transitioning between random detection and local siege. When |E|<1 and the smaller it is, the algorithm performs local siege; when |E|≥1 and the larger it is, the algorithm performs random detection; to balance global and local searches with dynamic energy, firstly, drawing on the idea of ​​hyperbolic tangent function, a dynamic hyperbolic tangent rate S is introduced at the position update in the early and middle stages of the algorithm iteration, and a sinusoidal fluctuation rate is introduced in the later stages of the iteration; the dynamic hyperbolic tangent rate used in the algorithm is used to improve the algorithm to obtain a larger prey energy in the early stages of the algorithm iteration to ensure that it searches for the optimal solution in a wider area; in the middle stages of the algorithm iteration, the algorithm is improved to obtain a smaller prey energy, so that the current individual can perform precise search in the optimal individual, effectively balancing global detection and local optimization; in the later stages of the algorithm iteration, the characteristics of irregular sine wave transformation are used to enhance the diversity of the optimal individual in the local space, and to assist the Harris hawk individual to jump out of the local optimum; Where S represents the dynamic hyperbolic tangent rate; S0 represents the initial velocity, t max represents the maximum number of iterations, t represents the number of iterations; S end Indicates the speed at the end of the iteration; when the number of iterations t = 0, S0 = 3, S end =0.4 indicates the speed at the end of the iteration, that is, the number of iterations reaches the maximum t max ; The smoothness of the curve is controlled by adjusting factors. When μ1=0.23,μ2=0.06,μ3=1.6,θ=0.3, the simulation results are optimal.

2. The AGV path planning method of the improved HHO algorithm fused with a neural network according to claim 1 is characterized by: The path smoothing process uses a Bezier curve, which can eliminate sharp turning points, making it more suitable for actual use of AGV.

3. The AGV path planning method of the improved HHO algorithm fused with a neural network according to claim 1 is characterized by: The environment is modeled using the visibility graph method, which uses convex polygons to fit the mapped obstacle information, thereby simplifying the path planning problem into the problem of finding the shortest route under constraints.

Images