Multi-strategy cooperative enhancement path planning method and navigation equipment

Abstract

The invention relates to the technical field of path planning, and provides a multi-strategy collaborative enhancement path planning method and navigation equipment. The path planning method comprises the following steps: acquiring an environment map comprising a starting position and a target position; initializing DBO algorithm parameters; initializing a dung beetle population based on Bernoulli chaotic mapping; calculating the fitness value of each dung beetle; determining the boundary of each dung beetle, and continuously iteratively updating the individual optimal fitness value and the global optimal fitness value of each dung beetle; and outputting the optimal path of the unmanned aerial vehicle until the number t of iterations is greater than the preset number T. According to the method, the initial solution is generated by introducing the chaos mapping based on Bernoulli distribution, so that the diversity and the ergodicity of the initial population are enhanced, premature convergence is effectively avoided, and the global search capability of the algorithm is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a path planning method and a navigation device with multi-strategy collaborative enhancement. Background Art

[0002] Path planning technology is the key to realizing the autonomous navigation of driverless vehicles. It requires the vehicle to be able to efficiently plan a route from the starting point to the end point based on a detailed surrounding environment model and ensure that all obstacles can be safely avoided during this process. Given the vastness of environmental data and the diversity of obstacles, the path planning problem has become a hot topic in current research.

[0003] For this reason, researchers have proposed many algorithms, such as the A* algorithm, Dijkstra algorithm, RRT algorithm, artificial potential field method, neural network algorithm, swarm intelligence algorithm, etc. The swarm intelligence algorithm operates by simulating the information exchange and cooperation mechanism among biological individuals, and realizes efficient intelligent performance at the swarm level through the optimization solution process of establishing a simple individual behavior mathematical model. Specifically, there are various swarm intelligence algorithms such as the particle swarm optimization algorithm (PSO), ant colony optimization algorithm (ACO), artificial bee colony algorithm (ABC), differential evolution algorithm (DE), sardine school algorithm, coati optimization algorithm (COA), grey wolf optimization algorithm (GWO), sparrow search algorithm, and dung beetle optimization algorithm (DBO).

[0004] The dung beetle optimization algorithm (DBO) is a swarm intelligence optimization algorithm that simulates the behavior of dung beetles rolling dung balls. The algorithm explores the search space by simulating the behavior of dung beetles during foraging, reproduction, dancing, etc. The DBO algorithm can be referred to in Chinese patents CN118349008A, CN117806346A, and CN118583163A. However, the existing DBO algorithm is prone to premature convergence, that is, it converges to the local optimal solution too early during the optimization process and fails to fully explore the entire search space, resulting in limited global search ability. Summary of the Invention

[0005] The purpose of the present invention is to provide a path planning method and a navigation device with multi-strategy collaborative enhancement, aiming to solve the technical problem that the existing DBO algorithm is prone to premature convergence.

[0006] In a first aspect, the present application provides a path planning method with multi-strategy collaborative enhancement, including the following steps:

[0007] S100: Obtain an environmental map including the starting position and the target position;

[0008] S200: Initialize the parameters of the DBO algorithm;

[0009] S300: Initialize the population of dung beetles based on the Bernoulli chaotic map, and its function expression is as follows:

[0010]

[0011] where x t represents the position of the dung beetle in the t-th iteration, d is a control parameter used to distinguish the coverage range of each partition space, and d ∈ (0, 1);

[0012] S400: Calculate the fitness value of each said dung beetle;

[0013] S500: Determine the boundary of each said dung beetle, and continuously iterate and update the individual best fitness value and the global best fitness value of each said dung beetle;

[0014] S600: Return to step S400 until the iteration number t is greater than the preset number T, and output the best path of the drone.

[0015] In a second aspect, the present application provides a navigation device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the multi-strategy collaborative enhanced path planning method as described in any one of the above.

[0016] The beneficial effects of the multi-strategy collaborative enhanced path planning method and navigation device provided by the present invention are as follows: In the environmental map, the DBO algorithm is used to plan the best obstacle avoidance path from the starting position to the target position. By introducing a chaotic map based on the Bernoulli distribution, an initial population with ergodicity and randomness is generated, effectively avoiding premature convergence and improving the global search ability of the algorithm in complex environments; by continuously updating the individual best fitness value and the global best fitness value of the dung beetle and controlling the number of iterations, the best path of the drone with high quality can be output in a short time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is the first flow schematic diagram of the path planning method provided by the embodiment of the present invention;

[0019] Figure 2 is the histogram and spatial scatter plot based on the Bernoulli chaotic map;

[0020] Figure 3 Schematic diagram of the process for obtaining an environmental map in the embodiment;

[0021] Figure 4 Schematic diagram of the process for obtaining the feature data of obstacles in the embodiment;

[0022] Figure 5 Schematic diagram of the structure of the laser beam splitter in the embodiment;

[0023] Figure 6 Second schematic diagram of the process of the path planning method provided by the embodiment of the present invention;

[0024] Figure 7 Variation curve graph of the crossover probability factor P;

[0025] Figure 8 Average iteration curve graph of the benchmark test function (dimension = 100);

[0026] Figure 9 Average iteration curve graph of the CEC2021 test function (dimension = 20);

[0027] Figure 10 Ranking result graph of the Friedman test;

[0028] Figure 11 Histogram of the path planning result;

[0029] Figure 12 Path planning graph with a map size of 10×10 and an obstacle ratio of 10%;

[0030] Figure 13 Path planning graph with a map size of 10×10 and an obstacle ratio of 20%;

[0031] Figure 14 Path planning graph with a map size of 10×10 and an obstacle ratio of 30%;

[0032] Figure 15 Path planning graph with a map size of 20×20 and an obstacle ratio of 10%;

[0033] Figure 16 Path planning graph with a map size of 20×20 and an obstacle ratio of 20%;

[0034] Figure 17 Path planning graph with a map size of 20×20 and an obstacle ratio of 30%;

[0035] Figure 18 Path planning graph with a map size of 30×30 and an obstacle ratio of 10%;

[0036] Figure 19 It is a path planning diagram with a map size of 30×30 and an obstacle ratio of 20%.

[0037] Figure 20 It is a path planning diagram with a map size of 30×30 and an obstacle ratio of 30%.

[0038] Figure 21 It is a schematic structural diagram of the navigation device provided by the embodiment of the present invention. Detailed implementation manners

[0039] The existing DBO (dung beetle algorithm) is a heuristic optimization algorithm based on the behavior of dung beetles, and its inspiration comes from the social behavior of the dung beetle population, including their behaviors of rolling balls, dancing, foraging, reproducing, and stealing.

[0040] I. Ball-rolling dung beetles

[0041] In the case of no obstacles, the dung beetle will move in a given direction within the entire search space. During the movement, the ball-rolling dung beetle not only explores the global space but also synchronously updates its position. The position update can be seen in Formulas (1) and (2).

[0042]

[0043] Among them, t represents the current iteration number; k is the deflection coefficient, and its value range is (0, 0.2]; α is the natural coefficient, and its value is 1 or -1. α = 1 means no deviation, and α = -1 means deviation from the original direction; b is a constant, and its value range is (0, 1); optionally, b = 0.3; ΔX is used to simulate the change in light intensity; X w represents the worst position in the current population; X i t represents the position of the i-th dung beetle in the t-th iteration; Lb and Ub respectively represent the lower and upper limits of the optimization problem.

[0044] II. Dancing dung beetles

[0045] When the dung beetle encounters an obstacle while rolling the dung ball, they will adjust their positions and search for new paths by dancing. At this time, the position update formula is Formula (3).

[0046]

[0047] Among them, θ represents the deflection angle, and its value range is [0, π]. When the value of θ is 0, π / 2, or π, the position of the dung beetle remains unchanged.

[0048] III. Reproducing dung beetles

[0049] Female dung beetles will roll and bury dung balls to create a safe environment for laying eggs. The equation for the egg-laying area of female dung beetles is:

[0050] Lb * =max(X * *(1 - R), Lb) (4)

[0051] Ub * =min(X * *(1 + R), Ub) (5)

[0052] Where Lb* and Ub* represent the lower and upper boundaries of the egg-laying area respectively, and they are dynamically adjusted with each iteration. X* represents the global ideal position of the population. R = 1 - t / T, where T is the maximum number of iterations, i.e., the preset number.

[0053] Each female dung beetle will produce a brood ball in each iteration and lay eggs within the egg-laying area. The following equation will dynamically adjust the position of the brood ball according to the change of the egg-laying area during the iteration process:

[0054]

[0055] Where represents the position of the i-th larva in the t-th iteration, b 1 and b 2 represent two independent random vectors of size 1×D, and D represents the dimension of the optimization problem.

[0056] IV. Foraging Dung Beetles

[0057] Equations (7) and (8) update the best foraging areas of adult dung beetles, which are the positions where they emerge from the ground.

[0058] Lb b =max(X b *(1 - R), Lb) (7)

[0059] Ub b =min(X b *(1 + R), Ub) (8)

[0060] Where X b represents the local optimal position of the current population; Lb b and Ub b are the lower and upper limits of the best foraging area respectively.

[0061] The position of the small dung beetle is updated as follows:

[0062]

[0063] Where C1 represents a random number subject to a normal distribution, C 2 represents a random vector between 0 and 1.

[0064] V. Theft of Dung Beetles

[0065] Some dung beetles steal dung balls from other dung beetles. These "thief" dung beetles update (i.e., search for or make new) dung balls around the optimal foraging sites:

[0066]

[0067] where g represents a random vector of size 1×D subject to a normal distribution, and S is a constant.

[0068] The first aspect

[0069] Combined Figure 1 , this application provides a path planning method with multi-strategy collaborative enhancement, including the following steps:

[0070] S100: Obtain an environmental map including the starting position and the target position.

[0071] S200: Initialize the parameters of the existing DBO algorithm.

[0072] S300: Based on the Bernoulli chaotic map, initialize the dung beetle population, and its function expression is as follows:

[0073]

[0074] where x t represents the position of the dung beetle in the t-th iteration, and d is a control parameter used to distinguish the coverage range of each divided space, d ∈ (0, 1). The original DBO algorithm cannot guarantee the uniform distribution of individuals at the starting positions in the search area, thus affecting the optimization performance of the algorithm. The chaotic map is a method for generating chaotic sequences, which is characterized by high randomness and ergodicity and has achieved better optimization results than random numbers. By using the chaotic map based on the Bernoulli distribution to generate the initial solution, the ergodicity and diversity of the initial dung beetle population can be increased, thereby improving the quality of the initial solution of the algorithm. Optionally, d takes the value of 0.518.

[0075] Figure 2 The left and right figures in Figure 2 respectively show the histogram and the spatial scatter distribution under chaotic conditions. Figure 2The scatter plot in it is used to show the relationship between the chaotic sequence (ordinate) and the number of iterations (abscissa), indicating that the chaotic sequence is evenly distributed at different numbers of iterations. Therefore, the initialization of the dung beetle population using the Bernoulli chaotic map can make the population more random and evenly distributed, improve the global search ability, reduce the impact of uneven initial values, and basically expand the spatial dispersion of the initial population.

[0076] S400: Calculate the fitness value of each dung beetle. Specifically, the cost function of the drone is the flight range cost of the path, and the fitness value of the dung beetle is calculated based on the dung beetle path data and the cost function of the drone. The flight range cost of the path is based on the trajectory length from the starting position to the target position.

[0077] S500: Determine the boundary of each dung beetle, and continuously iterate to update the individual best fitness value and the global best fitness value of each dung beetle.

[0078] S600: Return to step S400 until the number of iterations t is greater than the preset number T, and output the best path of the drone.

[0079] In the environmental map, the DBO algorithm is used to plan the best obstacle avoidance path from the starting position to the target position. By introducing a chaotic map based on the Bernoulli distribution, an initial population with ergodicity and randomness is generated, effectively avoiding premature convergence, improving the global search ability of the algorithm in complex environments, and solving the technical problem that the existing DBO algorithm is prone to premature convergence. By continuously updating the individual best fitness value and the global best fitness value of the dung beetles and controlling the number of iterations, the best path of the drone with high quality can be output in a short time.

[0080] In step S100 of the above embodiments, the source of the environmental map can be a map drawn from on-site shooting, purchased from a commercial navigation company, or a three-dimensional map taken by satellite, which is not uniquely limited here.

[0081] In some embodiments, in combination with Figure 3 , step S100 includes the following steps:

[0082] S110: Obtain the floor plan corresponding to the required environmental map. The floor plan can be a photographed picture, a planar map, which is not uniquely limited here.

[0083] S120: Perform vectorization processing on the floor plan to generate a two-dimensional vector map. The two-dimensional vector map is a two-dimensional graphic representation composed of a series of lines connected by points. The two-dimensional vector map is unified into a standardized vector format through vectorization processing. On the one hand, it is compatible with the GIS interface of the mainstream drone navigation system, reducing the data conversion cost. On the other hand, it is convenient for iteration and fitness value calculation in the dung beetle algorithm.

[0084] Optionally, the boundary of the map is converted into a digital form through image recognition technology.

[0085] S130: Obtain the characteristic data of the obstacle and identify it on the two-dimensional vector map to generate a three-dimensional environmental map. By obtaining and identifying the characteristic data of the obstacle, the three-dimensional map can more comprehensively reflect the potential risk points in the UAV flight environment.

[0086] Based on this, the obtained three-dimensional map provides richer spatial information than the two-dimensional vector map, including terrain undulation, building height, etc., which helps the UAV to more accurately plan the flight path and avoid obstacles.

[0087] In one embodiment, step S120 specifically includes the following steps:

[0088] S121: Obtain the shape features and geographical features of the floor plan.

[0089] S122: Perform data processing on the shape features and geographical features to generate a two-dimensional vector map.

[0090] Among them, the shape features generate information describing its geometric shape, and the geometric shape information may cover the coordinates of points, lines, and planes, etc. The geographical features are the attribute information of the two-dimensional vector map, such as the height and type of buildings, the width and grade of roads, etc. Data processing refers to encoding the shape features and geographical features to obtain feature data presented in the form of vectors, matrices, or tensors, etc.

[0091] In one embodiment, combined with Figure 4 , in step S130, the characteristic data of the obstacle is obtained through the following steps:

[0092] S131: Transmit multiple laser detection signals with different frequencies. Specifically, first, the control module sends a modulation signal to the drive module. Second, the drive module combines the modulation signal with the necessary DC bias current, and after appropriate amplification or adjustment, forms a modulation current with the correct bias and modulation amplitude, and transports the modulation current to the single longitudinal mode laser. Third, the single longitudinal mode laser drives the laser diode inside it to generate a single-frequency laser signal; under the action of the modulation current, the frequency of the laser signal will periodically change within a preset bandwidth to form a linear chirped optical signal. Finally, the laser beam splitter divides the linear chirped optical signal into a laser detection signal and a local oscillator signal. The local oscillator signal is transported to the mixer of the system for subsequent echo signal demodulation, and the laser detection signal is directly emitted towards the obstacle.

[0093] In one example, combined with Figure 5, the laser beam splitter includes a first coupling device 11, a phase adjuster 12, and a second coupling device 13 that are optically connected in sequence. The first coupling device 11 is responsible for splitting the linearly chirped optical signal into two beams of light - the first beam and the second beam. At the second coupling device 13, these two beams of light are further divided into a laser detection signal and a local oscillator signal. The role of the phase adjuster 12 is to adjust the phase of the first beam of light, thereby changing the phase difference between the two beams of light, allowing the second coupling device 13 to flexibly adjust the power distribution ratio between the laser detection signal and the local oscillator signal, increasing the optical power ratio of the laser detection signal, significantly reducing system losses, improving the utilization efficiency of light energy, and without relying on the configuration of an optical circulator.

[0094] Specifically, the splitting ratio can be dynamically adjusted within the range of 90:10 to 99:1. It can not only enhance the intensity of the laser detection signal to prevent the echo signal from being masked by the strong local oscillator signal, but also avoid unnecessary attenuation or discarding of the local oscillator signal, thereby optimizing the utilization of optical power. If there is a large deviation between the actual splitting ratio and the ideal value due to manufacturing process limitations, fine-tuning can be performed through the phase adjuster 12 to ensure that the output power of the actual local oscillator signal is consistent with or close to the output power in the ideal state. For example, when the optical power ratio of the laser detection signal to the local oscillator signal is set to 99:1, the laser detection signal exhibits excellent detection performance. At the same time, the intensities of the echo signal and the local oscillator signal are similar, which is conducive to the interference between the two to generate a clear interference signal, thereby simplifying the subsequent analysis process of the measurement results.

[0095] S132: Receive echo signals of different frequencies through multiple silicon photomultipliers. For example, multiple silicon photomultipliers are arranged in an M*M detector array, and each silicon photomultiplier is specifically configured to capture a specific, pre-set echo signal frequency. No two silicon photomultipliers will share the same echo signal frequency, ensuring that the echo signal frequencies detected by each silicon photomultiplier are different.

[0096] S133: Process the echo signals to generate characteristic data. First, use a phase shifter to perform phase distortion compensation operations on echo signals of different frequencies, aiming to correct the phase errors that may be introduced during signal propagation. Subsequently, through a beam combiner, the echo signals that have undergone phase distortion compensation are integrated and converged for light waves, accurately calculating the distance to the obstacle, and then inferring the specific shape of the obstacle and its position in space based on these distance data.

[0097] In some embodiments, in combination with Figure 6 , after step S400, the following steps are further included:

[0098] S410: Calculate the standard deviation of the best fitness values in five consecutive iterations.

[0099] S420: When the standard deviation is greater than the first preset value, if there is no obstacle, update the position of the dung beetle according to Equations (1) and (2).

[0100] S430: When the standard deviation is greater than the first preset value, if an obstacle is encountered, update the position of the dung beetle according to Equation (3).

[0101] Based on this, in five consecutive iterations, by monitoring the standard deviation, when the standard deviation is greater than the first preset value, it indicates that the algorithm is still continuously finding better solutions within the local search area. When the algorithm performs well in local search (i.e., the standard deviation is greater than the first preset value), continuing the search according to the current strategy can more effectively utilize computing resources, avoid unnecessary algorithm interference or reset, thereby improving the stability and efficiency of the overall search process.

[0102] Optionally, the first preset value ∈ [0, 1]. For example, the first preset value is 0.

[0103] In one embodiment, in combination with Figure 6 , after step S400, the following steps are further included:

[0104] S440: If the standard deviation is less than or equal to the first preset value, update the position of the dung beetle according to Equation (12):

[0105] X(t) = {0, c[2r(1) - 1], c[2r(2) - 1],..., c[2r(n) - 1]} (12)

[0106] where X(t) represents the set of random roaming step lengths, c is the cumulative sum of the roaming step lengths, n represents the maximum number of roaming times. r(t) is a random function, rand ∈ [0, 1], when the random number rand > 0.5, r(t) = 1; otherwise r(t) = 0.

[0107] By monitoring the standard deviation, if the best value of the dung beetle remains unchanged or changes little in five consecutive rounds, the random walk method is adopted to break the behavior of rolling the dung ball through Equation (12). Based on this, in the initial iteration stage, the range of random walk is large, which helps to enhance the global search ability. As the iteration progresses, the range of random walk gradually decreases, which helps to improve the ability of local search near the ideal position. In this way, the random walk strategy is introduced in the stage of the dung beetle rolling the ball behavior, enhancing the global optimization ability of the algorithm provided in this application and preventing it from falling into local minima. By randomly selecting a neighboring point of the current solution for comparison. If it performs better than the existing solution, it will replace the existing solution as the new solution. If a better value is not obtained after n consecutive attempts, it is considered that the optimal solution is located within an n-dimensional sphere, the center of which is the current optimal solution, and the radius is determined by the current step length.

[0108] In one embodiment, after step S500, the following steps are further included:

[0109] S510: The position of each dung beetle affects the boundary of its random walk. If the random number rand is greater than the crossover probability factor P, the boundary is updated based on the following formula:

[0110]

[0111] where, Ub t d and Lb t d respectively represent the upper and lower bounds of the d-th dimension in the t-th iteration, P t d represents the parameter value of the d-th dimension in the t-th iteration; I is a constant determined by the iteration number t and the maximum roaming number n.

[0112] Specifically, when 0.1 ≤ t / n < 0.5, I = 1 + 10 2 *t / n; when 0.5 ≤ t / n < 0.75, I = 1 + 10 3 *t / n; when 0.75 ≤ t / n < 0.9, I = 1 + 10 4 *t / n; when 0.9 ≤ t / n < 0.95, I = 1 + 10 5 *t / n; when 0.95 ≤ t / n < 1, I = 1 + 10 6 *t / n; when t / n is other values, I = 1.

[0113] Since the search range of the dung beetle is restricted within a certain boundary, it is necessary to normalize Equation (12) to obtain Equation (15).

[0114]

[0115] where, a i and b i respectively represent the minimum and maximum values of the i-th dung beetle's random walk in the d-th dimension; represents the position of the i-th dung beetle in the d-th dimension in the t-th iteration.

[0116] In some embodiments, in step S500, a horizontal crossover algorithm and a vertical crossover algorithm are used to calculate the individual best fitness value and the global best fitness value. The horizontal crossover algorithm and the vertical crossover algorithm ensure high convergence accuracy and fast calculation speed. In the population, the crossover operation performed on two individuals in the same dimension is called horizontal crossover, which enables cross-learning among numerous dung beetles, reduces search blind spots, and enhances the global search ability. Vertical crossover means performing a crossover operation on the optimal individuals from two different dimensions to increase the diversity of the optimal values.

[0117] Specifically, in the horizontal crossover algorithm, the parent dung beetles within the same dimension are randomly paired, and then crossover operations are performed between the parent individuals within each group to generate offspring. The offspring generated through horizontal crossover are compared with their parents in terms of fitness to retain the individuals with higher fitness, as shown in Equation (16).

[0118]

[0119] where a1 and a2 are arbitrary numbers between 0 and 1, while b1 and b2 are between -1 and 1; and respectively represent the parent individuals of the t-th generation in the d-th dimension after combination; and respectively represent the offspring individuals of the t-th generation in the d-th dimension generated through horizontal crossover from and

[0120] The following uses a specific numerical example to illustrate the horizontal crossover algorithm. The optimization problem is set to minimize the objective function f(x). This problem has two dimensions, and the population size is set to 10. The objective function is defined as:

[0121]

[0122] Initially, the population is generated by randomly selecting 10 points within the range [-5, 5], and their specific positions and corresponding objective function values are shown in Table 1. Subsequently, the population is randomly paired, and Table 2 summarizes the pairing results.

[0123] Values <![CDATA[X 1 > <![CDATA[X 2 > <![CDATA[X 3 > <![CDATA[X 4 > <![CDATA[X 5 > <![CDATA[X 6 > <![CDATA[X 7 > <![CDATA[X 8 > <![CDATA[X 9 > <![CDATA[X 10 > <![CDATA[x 1 > 0.6 2.5 4.7 1.6 -3.2 1.4 -0.4 0.7 1.8 -1.9 <![CDATA[x 2 > 3.8 -2.1 -1.9 -4.1 3.1 -3.6 3.3 2.2 4.6 -3.7 f(x) 15.16 16.91 47.79 21.93 30.09 16.88 11.21 5.82 27.64 20.91

[0124] Table 1 Initial positions and corresponding objective function values

[0125]

[0126] Table 2 Results of random pairing

[0127] The horizontal crossover algorithm is applied to Parent 1 and Parent 2. Taking the first pair as an example, for the first dimension of the offspring, random coefficient values are generated: a1 = 0.0216, a2 = 0.8440, b1 = -0.2656, b2 = 0.5682, and then the first dimension of the offspring is calculated as follows:

[0128]

[0129] where represents the first dimension of the offspring generated by Parent 1, represents the first dimension of the offspring generated by Parent 2.

[0130] For the second dimension of the offspring, random coefficient values are generated: a1 = 0.2254, a2 = 0.1735, b1 = 0.2707, b2 = -0.4697; then the second dimension of the offspring is calculated as follows:

[0131]

[0132] where represents the second dimension of the offspring generated by Parent 1, represents the second dimension of the offspring generated by Parent 2.

[0133] As a result of the horizontal crossover operation, Parent 1 represented by point X 1 generated Offspring 1 with coordinates (1.60, 0.03), while Parent 2 represented by point X 6 generated Offspring 2 with coordinates (2.57, 5.99). The other pairs of parents also produced their offspring in the same way. The offspring generated by all parents through horizontal crossover and their corresponding objective function values are shown in Table 3. The objective function values of the offspring were compared with those of their respective parents. For each pair, the solution with the lower objective function value was selected for position update after horizontal crossover. The results of the position update are summarized in Table 4.

[0134]

[0135] Table 3 Offspring Generated by Horizontal Crossover and Their Corresponding Objective Function Values

[0136] Values <![CDATA[X 1 > <![CDATA[X 2 > <![CDATA[X 3 > <![CDATA[X 4 > <![CDATA[X 5 > <![CDATA[X 6 > <![CDATA[X 7 > <![CDATA[X 8 > <![CDATA[X 9 > <![CDATA[X 10 > <![CDATA[x 1 > 1.60 2.50 -2.51 1.60 -1.73 1.40 1.08 0.42 3.10 -1.90 <![CDATA[x 2 > 0.03 -2.10 -3.26 -4.10 0.29 -3.60 -1.40 0.76 1.54 -3.70 f(r) 5.12 16.91 23.23 21.93 6.07 16.88 4.29 0.93 21.59 20.91

[0137] Table 4 Results of Position Update after Horizontal Crossover and Corresponding Objective Function Values

[0138] Specifically, to reduce the likelihood that certain dung beetle individuals in the population become trapped in local optima in specific dimensions during subsequent iterations, a vertical crossover algorithm is introduced. Vertical crossover is a mathematical crossover method that occurs between the dimensions of the favorable solutions generated by horizontal crossover. Each solution undergoes one vertical crossover, updating only one dimension while keeping the other dimensions unchanged. This can free the dimension trapped in the local optimum while retaining as much information as possible in other normal dimensions. Referring to Equation (17), the offspring generated by vertical crossover and their parents are evaluated, and only the individuals with better fitness are retained.

[0139]

[0140] where a3 is an arbitrary number between 0 and 1; and represent the global optimal solutions of the d1-th and d2-th dimensions in the t-th generation respectively; represents the offspring generated by vertical crossover.

[0141] Specifically, the crossover probability factor P is calculated as follows:

[0142]

[0143] where t represents the current iteration number and N represents the total number of iterations. If the random number generated in the 10th iteration is greater than the crossover probability factor P, then this crossover strategy will be used to fine-tune the optimal position of the current dung beetle. Otherwise, this strategy will be skipped.

[0144] Figure 7 Shows that the crossover probability factor P shows a non-linear downward trend as the number of iterations increases. To minimize the computational cost, in the early stage of the search, when the population diversity is large, the likelihood of using vertical and horizontal crossover methods will decrease. As the search progresses and the population diversity decreases, the probability of performing vertical and horizontal crossover increases, reducing the likelihood of converging to local optima and improving the global search ability of the algorithm.

[0145] Next, simulation experiments are carried out to prove the effectiveness of the proposed RCDBO algorithm. RCDBO refers to an algorithm generated by combining the existing DBO algorithm with the Bernoulli chaotic map, random walk, and vertical-horizontal crossover strategy. These experiments include testing 12 benchmark functions (dimensions 30, 50, 100) in the CEC2021 suite and the Bias Shift and Rotated Group functions (dimension 20). The selected comparative algorithms are some representative advanced swarm intelligence algorithms, including: Whale Optimization Algorithm (WOA algorithm), Subtraction Averaging-based Optimization Algorithm (SABO algorithm), Parrot Optimization Algorithm (PO algorithm), and the original Dung Beetle Optimization Algorithm (DBO), etc. At the same time, ablation experiment analysis, exploration and exploitation behavior analysis were carried out using the Bias Shift and Rotated Group functions in the CEC2021 suite, and the statistically significant Wilcoxon Rank-Sum Test and Friedman Test were conducted.

[0146] Each algorithm is executed with consistent parameters: the population size is set to 30, and other relevant parameters are not random numbers. See Table 5 for details. To ensure robustness, each algorithm is run 50 independent experiments to reduce accidental errors. The simulation software can be MATLAB R2021b or others, which is not uniquely defined here.

[0147] Algorithms Parameters RCDBO e = 0.4, Disturbance coefficient: k = 0.1, Luminous efficacy: b = 0.3 DBO Disruption factor: k = 0.1, Luminous efficacy: b = 0.3 IDBO <![CDATA[ScaleFactor: k = 10,000, Weight Factor: ω min = 0.2, ω max = 0.9, Correction coefficient: a = 0.45, Proportionality Constant: k p = 0.4]]> NCHHO Initial chaotic map parameter: a = 4 wOA Logarithmic spiral shape constant: b = 1, Convergence parameter a: Linear reduction from 2 to 0

[0148] Table 5 Parameter Settings

[0149] Table 6 shows the performance of 12 benchmark functions. Among them, f1 to f7 are unimodal functions, and f8 to f12 are multimodal functions. Unimodal functions are mainly used to test the convergence speed and accuracy of the algorithm; in contrast, multimodal functions, due to the existence of numerous local extrema, are used to test the ability of the algorithm to jump out of local optima and perform global search. Table 6 presents the mean, standard deviation, and minimum value of each method on 12 benchmark functions in the cases of dimensions 30, 50, and 100.

[0150]

[0151]

[0152] Table 6 Performance of 12 Benchmark Functions

[0153] Combined with Table 6, the results of 30 dimensions, 50 dimensions, and 100 dimensions show similarities, and the ranking of the ideal values has only changed slightly. In the tests for functions F1 to F4, F6, and F11, RCDBO achieved the theoretical ideal value of 0. For functions F8 to F10, although the test data of RCDBO may not always yield the theoretical ideal value of 0, it demonstrated significant advantages compared to other algorithms and always ranked first. Even in the case of function F5, although the average value and standard deviation of RCDBO were not the best, it was still superior to other algorithms in locating the minimum value.

[0154] Figure 8 It shows the average fitness iteration convergence comparison curves in these 50 independent experiments in the 100 - dimensional case. Except for functions F5 and F7, RCDBO showed higher convergence accuracy and speed on all functions. However, when testing these two functions, its stability was slightly inferior to the NCHHO and PO algorithms, resulting in the average convergence curve not being the best. Nevertheless, RCDBO still ranked first in solving the minimum value problem of F5. Therefore, RCDBO has more significant competitive advantages in dealing with unimodal and multimodal problems.

[0155] To comprehensively evaluate the optimization ability and convergence performance of RCDBO, the scope and difficulty of function testing were increased. An intelligent algorithm was used to handle the Bias shift and rot group functions in the more challenging CEC2021 test set, and comparative experiments on optimization accuracy and convergence curves were conducted. The CEC2021 test set contains ten benchmark functions, and the Bias shift and rot group applied bias, translation, and rotation transformations to these functions to increase the solution difficulty while reducing zero - tendency. The detailed function information is shown in Table 6. The test dimension of all functions was set to a maximum of 20 dimensions.

[0156] Table 7 shows the test results, where the optimal value of each item is highlighted in bold. Combined with Table 7, RCDBO achieved good optimization accuracy on most functions. In terms of the optimal value, average value, and standard deviation, RCDBO was superior to the other six comparison algorithms on functions CEC - 1, CEC - 4, and CEC - 7 to CEC - 10. In addition, RCDBO successfully identified the theoretical ideal value of the CEC - 6 function. Although the optimal value of RCDBO on the test function CEC - 2 was slightly lower than that of the traditional DBO algorithm, its average value was the best among all algorithms. For the CEC - 3 function, RCDBO had the optimal value, even though its average value and variance were inferior to the IDBO and NCHHO algorithms respectively. As for the CEC - 5 function, the average value and variance of RCDBO were both the best, indicating that its stability was superior to the other six algorithms.

[0157] In addition, Figure 9The average iteration curve in shows that when solving the average value of the CEC-3 function, the performance of RCDBO is slightly inferior to that of the IDBO algorithm. However, in terms of the accuracy of finding the optimal solution and the convergence speed of all other functions, RCDBO is significantly better than the other six algorithms. In short, compared with the other six comparison algorithms, RCDBO has better optimization accuracy and stability when optimizing the 20-dimensional Bias shift and rot group functions in the CEC2021 test set. This shows that RCDBO can still optimize the more challenging CEC2021 test set.

[0158] The Wilcoxon Rank-Sum Test determines its significance by calculating the p-value of the statistical results. The p-value is calculated by comparing RCDBO with the other six algorithms; if p < 0.05, it indicates that there is a significant difference between the two methods. Table 8 shows the results of the Wilcoxon Rank-Sum Test, and the parts with p-values greater than 0.05 are highlighted. The last row in the table indicates the number of cases where the performance of RCDBO is better than, equal to, or inferior to that of the other algorithms by the symbols + / = / - From the fact that most of the p-values in the table are below 0.05, it can be seen that in the ten benchmark functions, there is a significant difference in the optimization performance between RCDBO and the other six algorithms, and RCDBO performs better.

[0159]

[0160] Table 7 Test Results of the CEC2021 Test Set

[0161] Fun No. DBO IDBO NCHHO SABO WOA PO CEC - 1 5.02E-17 3.19E-07 7.07E-18 7.07E-18 7.07E-18 7.07E-18 CEC - 2 1.47E-05 4.26E-01 7.07E-18 7.07E-18 3.02E-15 4.90E-13 CEC - 3 2.06f-01 7.80E-01 7.07E-18 1.10E-13 1.54E-17 1.17E-15 CEC - 4 5.29E-14 1.98E-07 7.07E-18 7.97E-18 7.97E-18 1.54E-17 CEC - 5 2.89E-10 8.71E-03 8.99E-18 2.05E-09 2.56E-11 1.12E-10 CEC - 6 7.97E-04 6.32E-02 1.01E-17 1.45E-17 7.49E-16 4.93E-09 CEC - 7 7.80E-12 2.22E-11 8.46E-18 4.50E-16 3.38E-15 4.36E-09 CEC - 8 2.20E-17 1.29E-10 7.07E-18 7.07E-18 7.07E-18 7.07E-18 CEC - 9 5.56E-12 1.93E-06 7.07E-18 3.38E-15 2.54E-16 2.30E-08 CEC - 10 Z12E - 01 1.65E-01 7.07E-18 7.07E-18 3.53E-17 9.92E-16 + / - / ＝ 8 / 2 / 0 6 / 4 / 0 10 / 0 / 0 10 / 0 / 0 10 / 0 / 0 10 / 0 / 0

[0162] Table 8 Results of the Wilcoxon Rank-Sum Test

[0163] The Friedman test ranks the algorithms according to the calculated average rank, and a higher rank indicates better algorithm performance. The ranking results of the Friedman test are as Figure 10 shown. Except that the algorithm provided in this application ranks second in the CEC-3 function, it ranks first in all other functions.

[0164] RCDBO1, RCDBO2, and RCDBO3 are three algorithms generated by separately combining the existing DBO algorithm with Bernoulli chaotic mapping, random walk, and vertical-horizontal crossover strategy, namely three specific embodiments of the algorithms provided in this application. To evaluate the effectiveness and feasibility of these strategies, the RCDBO algorithm was compared with RCDBO1, RCDBO2, RCDBO3, and the original DBO algorithm on the Bias Shift and Rotated Group functions in the CEC2021 test set. Table 9 shows the comparison results, where the ideal values are shown in bold. Figure 9 It shows that RCDBO performs the best.

[0165]

[0166] Table 9 Results of ablation experiments

[0167] In the environmental map, the moving vehicle is regarded as a particle. Obstacles are represented by 1 and non-obstacles are represented by 0: where, (x k , y k ) represents the coordinates of the k-th node, d represents the length of the route that the vehicle travels from the starting point to the ending point, and n represents the number of intermediate points in the journey.

[0168] In environmental maps of three sizes, 10×10, 20×20, and 30×30, with the proportion of obstacles between 10% and 30%, nine different experimental environments were generated, simulating simple and complex obstacle scenarios. Each method was independently run 50 times under the condition that the maximum number of iterations was limited to 100 to reduce the influence of randomness.

[0169] Figure 11 And Tables 10, 11, and 12 show the comparison and results of path planning data. In each map scenario, the average path length and ideal path length of the RCDBO method are better than those of the DBO algorithm. Figures 12 to 20 It shows the iteration curves of the best and average paths designed by the RCDBO and DBO algorithms in environments of different sizes. In each experimental case, the initial path length value predicted by RCDBO is better than that of the DBO method. This indicates that by increasing population diversity and traversal ability, the RCDBO algorithm improves the quality of early solutions.

[0170] When the map size is 10×10, as Figure 12 , Figure 13 and Figure 14As shown, in the 10% and 20% environments with simple obstacle distributions, the paths planned by the RCDBO method are relatively smooth, even without turning points. In the more challenging 30% obstacle distribution scenario, although the path planned by RCDBO has more turning points, this is to achieve a shorter total path length. The analysis of the average iteration curve shows that in the settings with 10% and 20% obstacle distributions, the RCDBO algorithm converges faster than the DBO algorithm initially, indicating that its iteration speed has been optimized after increasing the population diversity. Although in the complex environment with 30% obstacle dispersion, RCDBO on average requires more iteration times than DBO, it continuously finds better solutions as the number of iterations increases. This shows that while RCDBO prevents premature convergence by disturbing the dung beetle population and increases the possibility of finding the global optimal solution, DBO is prone to falling into the local optimal trap throughout the iteration stage.

[0171] Compared with the 10×10 map, the 20×20 and 30×30 maps are more difficult to find solutions. Figures 15 to 20 Shows the optimal paths provided by the RCDBO and DBO algorithms for these two map sizes. In the 30×30 sized map environment with obstacle ratio distributions of 20% and 30%, in order to solve the shortest path, the RCDBO algorithm increases the number of turning points of the planned path compared to the DBO algorithm. However, in other cases, due to the relatively simple map environment, the optimal path generated by the RCDBO algorithm has fewer turning points, is shorter and smoother than the path generated by the DBO algorithm. From the average iteration curve, as the map size and complexity increase, the challenges also increase. DBO tends to converge prematurely and fall into local optima, while RCDBO shows a more obvious advantage in escaping these traps. RCDBO is always superior to DBO in deeper iterations, indicating that it strengthens its global optimization ability through strategies such as random walks, vertical and horizontal crossovers. However, it can also be seen that in the environment with dense obstacles, when the shortest path is the only goal of path planning, the turning points of the path planned by RCDBO increase, and since the algorithm combines multiple strategies, its computational cost also increases.

[0172] Compared with DBO, the optimization performance of RCDBO is significantly improved, with higher path planning efficiency and better results. Even in complex environments, RCDBO can avoid premature convergence and local minimum traps, demonstrating excellent global path planning capabilities.

[0173] In some embodiments, after step S600, the following steps are further included:

[0174] S710: Obtain a first intermediate point and a second intermediate point that are successively distributed on the optimal path. The first intermediate point and the second intermediate point divide the optimal path into a first path, a second path, and a third path, and the lengths of the first path, the second path, and the third path are not less than 1 / 4 of the length of the optimal path.

[0175] S720: Re-obtain the first path, the second path, and the third path respectively according to steps S200 to S600.

[0176] S730: Combine the first path, the second path, and the third path obtained in step S720 into the optimal path.

[0177] To further ensure that the optimal path is the shortest path, the optimal path is split into a first path, a second path, and a third path, and the RCDBO is respectively used for more detailed planning and optimization to avoid a certain section falling into a local optimal solution, which is a verification and redundant design of the original optimal path. This redundant design improves the reliability of path planning and reduces the overall failure risk caused by a single path problem.

[0178] In summary, it is known that the existing DBO algorithm is prone to falling into local optimality and has weak global search ability, and there is an imbalance between its local development and global search ability. To overcome these limitations, RCDBO is provided. The population initialization is carried out by using the Bernoulli-based chaotic mapping technology, which enhances the uniform randomness of the initial population, thereby improving the quality and efficiency of the first batch of solutions. Given the importance of the dung beetle's behavior of rolling dung balls in global optimization, when the current optimal solution has not changed for five consecutive iterations, a random walk strategy is implemented to disrupt the dung beetle's rolling behavior, aiming to improve the effectiveness of early optimization. At the same time, vertical and horizontal crossover methods are introduced, which have the risk of disturbing the optimal position of the current dung beetle population, aiming to prevent the algorithm from stagnating in the later stage. Therefore, RCDBO prevents the algorithm from falling into local optimality, improves its global optimization ability, and finds a balance between local exploration and global exploration.

[0179] In addition, in terms of finding the best route in a grid map, the enhanced RCDBO algorithm outperforms the original DBO method in terms of the most efficient manner and average path length in nine maps with different complexities and sizes. In the context of a simple obstacle map, the RCDBO method shows a faster iterative optimization speed and generates a smoother path. The RCDBO algorithm can navigate in a complex obstacle map environment by avoiding premature convergence and finding a more efficient global path than the original DBO method. The test results show that when these three methods are combined, the RCDBO method outperforms the DBO method in terms of performance. It performs excellently in dealing with test function problems and path planning problems in navigation applications.

[0180] In the second aspect, in combination withFigure 21 , this application provides a navigation device 20, including a memory 22, a processor 21, and a computer program 23 stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program 23, it implements the multi-strategy collaborative enhanced path planning method as described in any one of the above.

[0181] Among them, the navigation device 20 can be a mobile phone, a tablet computer, a wearable device, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc., which is not uniquely limited here.

[0182] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-strategy collaborative enhanced path planning method, characterized in that: The method comprises the following steps: S100: Acquire an environment map including a starting position and a target position; S200: Initialize DBO algorithm parameters; S300: Based on the Bernoulli chaos map, the dung beetle population is initialized, and its function expression is as follows: Among them, x t represents the position of the dung beetle in the tth iteration, d is a control parameter used to distinguish the coverage of each partition space, d∈(0,1); S400: Calculating the fitness value of each dung beetle; S500: determining the boundary of each dung beetle, and continuously iteratively updating the individual best fitness value and the global best fitness value of each dung beetle; S600: Return to step S400 until the number of iterations t is greater than the preset number T, and output the optimal path of the UAV.

2. The multi-strategy collaborative enhanced path planning method according to claim 1, characterized in that: After step S400, the following steps are also included: S410: Calculate the standard deviation of the best fitness value in five consecutive iterations; S420: When the standard deviation is greater than the first preset value, if there is no obstacle, the position of the dung beetle is updated according to the following formula: Where k is the deflection coefficient, with a range of (0, 0.2]; α is the natural coefficient, with a value of 1 or -1, α = 1 means no bias, α = -1 means deviation from the original direction; b is a constant, with a range of (0, 1); ΔX is used to simulate the change of light intensity, X w Represents the worst position in the current population; S430: When the standard deviation is greater than the first preset value, if an obstacle is encountered, the position of the dung beetle is updated according to the following formula: Where θ represents the deflection angle, and its value range is [0, π].

3. The multi-strategy collaborative enhanced path planning method according to claim 2, characterized in that: After step S400, the following steps are also included: S440: If the standard deviation is less than or equal to the first preset value, the position of the dung beetle is updated according to the following formula: X(t)={0, c[2r(1)-1], c[2r(2)-1],..., c[2r(n)-1]}; Among them, X(t) represents the set of random walk step lengths, c is the cumulative sum of walk step lengths, n represents the maximum number of walks, and r(t) is a random function.

4. The multi-strategy collaborative enhanced path planning method according to claim 3, characterized in that: After step S500, the following steps are also included: If the random number rand is greater than the crossover probability factor P, the boundary is updated based on the following formula: Among them, rand∈[0,1], Ub t d With Lb t d They represent the upper and lower bounds of the dth dimension in the tth iteration, respectively, t d represents the parameter value of the dth dimension in the tth iteration; I is a constant determined by the number of iterations t and the maximum number of roaming times n.

5. The multi-strategy collaborative enhanced path planning method according to claim 1, characterized in that: In step S500, the individual best fitness value and the global best fitness value are calculated using the horizontal crossover algorithm and the vertical crossover algorithm.

6. The multi-strategy collaborative enhanced path planning method according to claim 1, characterized in that: Step S100 includes the following steps: S110: Obtaining a plan view corresponding to the required environment map; S120: performing vectorization processing on the plane diagram to generate a two-dimensional vector diagram; S130: Obtain feature data of obstacles and mark them on the two-dimensional vector map to generate the three-dimensional environment map.

7. The multi-strategy collaborative enhanced path planning method according to claim 6, characterized in that: Step S120 specifically includes the following steps: S121: Acquire shape features and geographical features of the plan view; S122: Perform data processing on the shape feature and the geographic feature to generate a two-dimensional vector map.

8. The multi-strategy collaborative enhanced path planning method according to claim 6, characterized in that: The step of obtaining characteristic data of obstacles comprises the following steps: S131: transmitting a plurality of laser detection signals of different frequencies; S132: Receiving echo signals of different frequencies respectively through a plurality of silicon photomultiplier tubes; S133: Process the echo signal to generate the characteristic data.

9. The multi-strategy collaborative enhanced path planning method according to any one of claims 1 to 8, characterized in that: After step S600, the following steps are also included: S710: Acquire a first middle point and a second middle point distributed in sequence on the optimal path, wherein the first middle point and the second middle point divide the optimal path into a first path, a second path, and a third path, and the lengths of the first path, the second path, and the third path are not less than 1 / 4 of the length of the optimal path; S720: Reacquire the first path, the second path, and the third path respectively according to steps S200 to S600; S730: Combining the first path, the second path, and the third path obtained in step S720 into the optimal path.

10. A navigation device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the multi-strategy collaborative enhanced path planning method according to any one of claims 1 to 9 is implemented.

Images