Multi-aUV cooperation method based on artificial bee colony and artificial potential field in sea area search

Abstract

The application relates to a multi-AUV cooperation method based on artificial bee colony and artificial potential field in sea area search, which comprises the following steps: performing fine-grained and coarse-grained environment modeling on a task area, and dividing the area into feasible domains and infeasible domains. Then, an improved voronoi clustering strategy is adopted to reasonably divide the coarse-grained cubic model into several small areas, which are assigned to AUVs one by one. Then, an artificial bee colony algorithm is adopted to plan a global path for each AUV, and the global path is smoothed by adopting a line-of-sight technology. Meanwhile, considering the need of obstacle avoidance and collision avoidance, when an AUV detects an obstacle or other AUVs on the global path, an artificial potential field method is adopted to plan a local path for avoidance.

Description

Technical Field

[0001] This invention relates to a multi-AUV cooperative method based on artificial bee colonies and artificial potential fields in marine search, belonging to the field of underwater data collection technology. Background Technology

[0002] AUV search methods in marine environments have undergone some development, which can be divided into research on single AUVs and research on multi-AUV swarms, mainly including steps such as environmental modeling, global path planning, and local path planning. For single AUV missions, efficient real-time path planning and efficient area search are issues that deserve special attention; for multi-AUV swarm missions, communication methods, formation control, and cooperative operations are key research directions for the future.

[0003] Currently, path planning algorithms for UAVs in aerial environments and intelligent vehicles in terrestrial environments have been gradually perfected through extensive research by scholars both domestically and internationally. However, the path planning problem in underwater environments differs from that of terrestrial and aerial environments, which mostly use two-dimensional modeling methods. Underwater environments are complex, dynamic, three-dimensional environments filled with randomness and uncertainty, and are permeated with underwater natural environmental factors such as turbulence and eddies. Furthermore, underwater environments contain numerous complex obstacles. In addition to static environmental obstacles obtained by AUVs from static nautical charts, dynamic obstacles composed of unknown terrain and other static obstacles, as well as underwater fish schools and various other factors, also permeate the AUV's navigation path.

[0004] In environmental modeling, common methods include topology, grid methods, and visualization methods. Topology is complex and requires significant storage space, limiting its research progress. Since the marine environment is three-dimensional and ocean currents have a significant impact, grid and visualization methods are limited to two-dimensional planes, resulting in limited practicality. In path planning, common methods include the A* algorithm and its various improvements, the RRT algorithm and its various improvements, artificial potential field methods, swarm intelligence algorithms, reinforcement learning algorithms, and neural network algorithms. The artificial potential field method abstracts robot motion into motion within a virtual potential field. Obstacles exert a repulsive force on the robot, while targets exert an attractive force; the combined effect of these two forces guides the robot's movement. Swarm intelligence algorithms mimic biological social behavior, using cooperation mechanisms between individuals to exchange information and continuously update the optimal solution for the entire group until convergence to the global optimum. In swarm intelligence algorithms, the artificial bee colony algorithm is an optimization method that imitates the behavior of bees. Its main feature is that it does not require knowledge of specific information about the problem. It only needs to compare the merits of the problem and, through the local optimization behavior of each artificial bee, eventually makes the global optimum emerge in the swarm, resulting in a relatively fast convergence speed.

[0005] To achieve efficient and energy-saving searches in marine environments, many researchers both domestically and internationally have proposed numerous search schemes. Relevant literature is as follows:

[0006] 1. Zhu Daqi et al., in their paper "Biologically Inspired Self-Organizing Map Applied to Task Assignment and Path Planning of an AUV System," proposed a biomimetic self-organizing map algorithm, BISOM, for obstacle avoidance task allocation and path planning in a three-dimensional underwater environment for AUV systems. This algorithm embeds a biomimetic neural network into a self-organizing map neural network. The goal of task allocation and path planning is to schedule a group of AUVs to visit all designated target locations while ensuring obstacle avoidance without velocity jumps. The self-organizing map neural network is used to allocate a group of AUVs to reach multiple target locations in the underwater environment. Then, to ensure that each AUV reaching a target location avoids obstacles and velocity jumps, the weights of the winners of the self-organizing map are updated using the biomimetic neural network, achieving path planning and effective navigation for the AUVs. Although this method has superior performance, some issues, such as ocean currents, communication, and positioning, are still not fully considered.

[0007] 2. Cui Rongxin et al., in "Mutual Information-Based Multi-AUV Path Planning for Scalar Field Sampling Using Multidimensional RRT*", proposed a multi-AUV adaptive path planning algorithm for estimating the scalar field of a region of interest. This method divides the cooperative sampling of multiple AUVs into two steps: 1) distributively modeling the scalar field using data acquired from sensors; 2) planning a path for each AUV using the estimated model to improve the quality of the next sampling location. This paper proposes a Selective Basis Function Kalman Filter (SBF-KF) to estimate the scalar field model in a distributed manner using a consensus algorithm. In traditional path planning problems, a definite initial state and a target state are usually given. For scalar field sampling problems, only the initial state of the AUV is available. Each AUV must determine its next sampling location based on real-time sensor data. To find an available path that minimizes observation uncertainty, this paper uses mutual information to represent the uncertainty of the scalar field model. Based on this, this paper proposes a mutual information-based multidimensional RRT* algorithm, MDMI-RRT*, which simultaneously constructs multiple route maps, guiding AUVs to locations where more information can be collected. Furthermore, mutual information is used to represent sampling cost, unlike traditional path planning which uses path length or the time elapsed on the path.

[0008] 3. Zhang Lanyong et al., in "Three-Dimensional Underwater Path Planning Based on Modified Wolf Pack Algorithm," proposed an improved wolf pack algorithm. By improving three intelligent behaviors of the wolf pack algorithm—reconnaissance, summoning, and encirclement—they overcame the shortcomings of slow convergence speed and low convergence accuracy. To improve search behavior, interactive search was proposed to increase the interactivity among wolves. Furthermore, to improve summoning behavior, an adaptive step size model based on prey was established to enhance search capabilities. Finally, a new calculation rule for encirclement behavior was designed, significantly enhancing the ability for fine-grained local search. The artificial wolf can not only perceive information from its companions but also transmit information from its prey, thus effectively controlling the balance between global and local searches. In addition, the ambush step size was changed from a fixed value to an adaptive value to improve the algorithm's global search capability and prevent it from getting trapped in local optima. Furthermore, to meet the constraints of AUVs, a fast path planning method based on dubins paths was proposed. Dubins path planning is used to satisfy angle control constraints, and the turning radius is adjusted to meet these constraints. By establishing an underwater environmental threat model, a dubins path planning method based on the improved wolf pack algorithm was proposed. Without considering ocean currents, the path between waypoints is the shortest, posing the least threat and consuming the least energy. However, the improved wolf pack algorithm requires setting too many parameters, and the settings of some parameters significantly affect the convergence speed and the accuracy of the results. Inappropriate parameter settings may lead to low accuracy or convergence that is too fast or too slow.

[0009] 4. Stefano Basagni et al., in "Path Finding for Maximum Value of Information in Multi-Modal Underwater Wireless Sensor Networks," proposed a greedy adaptive AUV path planning heuristic algorithm, GAAP, which drives the AUV to collect data from nodes based on the information value of the node data. To benchmark the performance of the AUV path planning heuristic algorithm, an integer linear programming formula was defined that accurately simulates the considered scenario, yielding a path that drives the AUV to collect and transmit data with maximum information value. Then, a realistically deployable heuristic algorithm was defined for AUV path planning, adaptable to events occurring at unpredictable locations and times. The AUV selects the next node to visit based on the information value it expects to collect at the next location. The information needed to make this decision is propagated to the AUV via short event packets transmitted acoustically. The AUV plans to visit a node if and only if visiting a node that has already sent an event packet would increase the information value of the data transmitted by the AUV to the receiver.

[0010] 5. Narcís Palomeras et al., in "Autonomous Exploration of Complex Underwater Environments Using a Probabilistic Next-Best-View Planner," proposed a probabilistic next-best-view planner for hovering AUVs, allowing them to explore complex environments without prior models. This exploration results in a set of scans obtained within a reasonable timeframe, maximizing coverage of a defined area. Unlike some UAV methods where the UAV continuously collects data along a trajectory, the proposed method, similar to methods used for 3D object reconstruction, acquires scans only from static viewpoints to reduce drift accumulation. The proposed method consists of the following steps: First, several candidate viewpoints are generated from the AUV's current location and pruned according to a series of criteria. Then, a utility function (i.e., the expected increase in information) for each candidate viewpoint is evaluated using mapped sensor specifications. After probabilistically determining the viewpoint with the maximum utility, an unobstructed path to that viewpoint is planned, and the AUV is navigated to it. Once the viewpoint is reached, a new scan is collected and registered with previous scans. This registration can be used to improve the consistency of AUV positioning and environmental representation. Repeat the process until the termination criteria are met. Summary of the Invention

[0011] The technical problem this invention aims to solve is to perform fine-grained and coarse-grained environmental modeling of the task area, dividing the area into feasible and infeasible regions. Then, an improved Voronoi clustering strategy is used to rationally divide the coarse-grained cubic model into several smaller regions, which are then assigned to AUVs. Next, an artificial bee colony algorithm is used to plan a short global path with a small turning angle for each AUV, and line-of-sight techniques are employed to smooth the planned global paths. Simultaneously, considering obstacle avoidance and collision avoidance requirements, when an AUV detects an obstacle or other AUV on its global path, an artificial potential field method is used to plan a local path to avoid it.

[0012] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0013] A multi-AUV collaborative method based on artificial bee colonies and artificial potential fields in marine search includes the following steps:

[0014] Step 1: Environmental modeling of the mission area. Considering the uncertainty and time sensitivity of the marine environment, the environmental information of the mission area is unknown and dynamic. It is necessary to divide the area into feasible and infeasible regions. Furthermore, clustering and path planning algorithms require uniformly distributed target locations and waypoints. Therefore, coarse-grained and fine-grained cube modeling methods are used to model the environment of the mission area.

[0015] Step Two: Task Region Division and Allocation. Considering the large size of the task region, an improved Voronoi clustering strategy is used to divide it into several smaller regions, which are then assigned to AUVs, ensuring that each AUV is responsible for searching a small region. The center of the coarse-grained cube generated in Step One is used as the target location for the Voronoi clustering strategy. Each target location is assigned to the AUV that takes the shortest time to reach it. A number of assigned target locations and a continuity constraint are added to the calculation to avoid uneven distribution and ensure the continuity of the smaller regions.

[0016] Step 3: Global Path Planning. Considering the limited energy and maneuverability of the AUV, a short global path with a small turning angle needs to be planned in advance. Here, the Artificial Bee Colony algorithm is used, and line-of-sight technology is introduced to smooth the path planned by the algorithm. Since this invention targets unknown environments, i.e., without any prior knowledge, while the Artificial Bee Colony algorithm does not require prior knowledge and has a faster convergence speed.

[0017] Step 4: Local Path Planning. While navigating the global path planned in Step 3, the AUV will detect obstacles or other AUVs, i.e., infeasible regions. If the infeasible region is located on the global path, a local path needs to be planned to avoid it. Here, the artificial potential field method is used. This method has low computational cost and is robust to control and sensing errors, making it suitable for real-time local path planning.

[0018] In step one above, the environmental modeling method for the task area is as follows:

[0019] First, assume the task area is a cube, then divide it into smaller cubes. The side length of each smaller cube is set based on the AUV's detection range. In coarse-grained cube modeling, the center of all cubes is used as the target location for task area division and allocation. In fine-grained cube modeling, the center of each cube is used as a waypoint in path planning. For example, if the AUV's detection range is 30-50m, coarse-grained modeling would set the side length to 60m, ensuring the AUV can detect the entire cube from its center, thus using the center of all cubes as the target location. Fine-grained modeling would set the side length to 20m, using the center of each cube as a waypoint.

[0020] In step two above, the method for dividing and allocating task areas is as follows:

[0021] When using the improved Voronoi clustering strategy to partition and allocate task regions, a coarse-grained cubic model is employed, where the center of each cube is used as the target location. Assuming m AUVs are randomly distributed outside the task region, the Voronoi clustering strategy, considering the location of each AUV, is formulated as follows:

[0022]

[0023] Where p j It is the position of the j-th AUV, t(p j (k) represents the time required for the j-th AUV to travel from its location to the target k;

[0024] The improved Voronoi clustering strategy changes the objective function with respect to time to an objective function with respect to the distance between assigned and unassigned target locations, and adds a constraint on the number of assigned target locations;

[0025] The improved Voronoi clustering strategy can be formulated as follows:

[0026]

[0027] Where T j T is the set of target locations that have been assigned to the j-th AUV. u It is the set of unclustered target locations, and d(i,k) represents the Euclidean distance between target location i and target location k. j This represents the number of target locations already assigned to the j-th AUV. α and β are two parameters that can be set manually. This improvement has two advantages: first, it avoids some AUVs being assigned too many target locations, while others are assigned too few or none, resulting in an excessively long total search time; second, it allows the target locations assigned to each AUV to form a continuous search area.

[0028] In step three above, the global path planning method is as follows:

[0029] When using a combination of artificial bee colony algorithm and line-of-sight technique for global path planning, fine-grained cube modeling is employed, where the center of each cube serves as a path point. The artificial bee colony in the artificial bee colony algorithm is divided into three groups: mercenary bees, follower bees, and scout bees. For each journey segment, N paths are randomly generated, resulting in N solutions, which are also N nectar sources. Each solution x... i (i = 1, 2, ..., N) are represented by a d-dimensional vector x i =(x i,1 ,xi,2 ,…,x i,d Let d be the number of path points. At the beginning of the search, the hired bees search around nectar source i to generate a new nectar source according to the following formula:

[0030] x i,j '=x i,j +r i,j (x i,j -x k,j (3)

[0031] Where j is a random integer in [1, d], representing that the hired bee randomly selects one dimension for the search; k≠1 means randomly selecting a honey source that is not equal to i from N honey sources; r i,j The random number is uniformly distributed in the range [-1, 1], which determines the perturbation amplitude. The hired bee calculates the fitness values ​​of nectar source i and the new nectar source, and selects the better nectar source using a greedy algorithm. The following bee follows the nectar source based on the nectar source information shared by the hired bee, with the probability calculated using the following formula:

[0032]

[0033] Among them, fit i The fitness value of the i-th solution is calculated using the following formula:

[0034]

[0035] Among them, f i Let x be the objective function value. The objective of path planning is to minimize the length cost and turning cost of the path. i The length cost is defined as d i =d(x i,1 ,x i,2 )+d(x i,2 +x i,3 )+…+d(x i,d-1 ,x i,d ), where d(x) i,j, x i,j+1 ), It is the Euclidean distance between two consecutive path points; path point x i,j The turning cost at point t(x) is i,j ), Defined as follows: Let the initial position of the AUV be x. i,0 , with x i,j (j∈{1,2,…,d-1}) is the endpoint, passing through x i,j-1 The ray is l j,j-1 , with x i,j As the endpoint, passing through x i,j+1 The ray is l j,j+1Then t(x) i,j )=cos <l j,j-1 ,l j,j+1 For path x i ,make For all nonnegative t(x) i,j ), For all negative t(x) i,j The sum of ). Therefore,

[0036]

[0037] Here, γ and δ are two parameters that can be set manually. Then, the follower bee selects the hired bee using a roulette wheel method, that is, it generates a uniformly distributed random number r in [0,1]. If p i If the value is greater than r, the follower bee generates a new nectar source around nectar source i in the same way as the hired bee, and uses the same greedy selection method to determine which nectar source to retain. During the search process, if nectar source i reaches the threshold l after a finite number of iterations... limit If no better nectar source is found, the nectar source will be abandoned, and the corresponding hired bee will become a scout bee. The scout bee will randomly generate a new nectar source in the search space to replace nectar source i, that is:

[0038]

[0039] in, This is the minimum value obtained so far in the j-th dimension. To obtain the maximum value in the j-th dimension, record the best honey source to date.

[0040] Simultaneously, line-of-sight (LOS) technology is employed to smooth the planned global path: a LOS check is performed at each pathpoint on the global path to determine whether the pathpoint can be skipped, i.e., whether it is possible to directly reach the next pathpoint from the previous pathpoint. For the target location pathpoint, no LOS check is performed, meaning that this pathpoint cannot be skipped. The smoothed path is more suitable for practical AUV applications.

[0041] In step four above, the local path planning method is as follows:

[0042] When using the artificial potential field method for local path planning, fine-grained cube modeling is employed, where the center of each cube serves as a path point. When the AUV detects an infeasible region on the global path, the first path point before the infeasible region is designated as the starting point of the local path, and the first path point after the infeasible region is designated as the target point. Let the current path point of the AUV be X, and the target point be X₀. goal The gravitational field exerted by the target point on the AUV can be designed as follows:

[0043]

[0044] Where k is the position gain function, XX goal This is the distance between the AUV and the target point. The corresponding gravitational force is the negative gradient of the gravitational field function:

[0045]

[0046] During the AUV's journey to the target point, this force converges linearly to zero. The repulsive force field exerted by the infeasible region on the AUV can be designed as follows:

[0047]

[0048] Where η is the position gain function, ρ is the shortest distance between the AUV's position x and the obstacle, ρ0 is a constant representing the obstacle's influence distance, and n is an arbitrary constant greater than zero. The introduction of the distance between the AUV and the target ensures that the entire potential field exists only at the target point X. goal Global minimum. When X≠X goal When the repulsive force is at that time, it can be expressed as:

[0049]

[0050] in,

[0051]

[0052]

[0053]

[0054] Vector F re The direction is from the infeasible region to the AUV, vector F at The direction is from the AUV towards the target point. Clearly, when F... re When F generates a repulsive force on the AUV, at An attractive force is exerted on the AUV, directing it towards the target point. In local path planning, the direction of the vector from the current path point to the next path point needs to be as close as possible to the potential force F. sum If the direction is such that the objective function of local path planning is , then the objective function of local path planning can be designed as:

[0055]

[0056] Where Ω is the set of candidate path points for the AUV at the next time step, and θ j From the current path point to the candidate path point w j The angle between the vector and the direction of the potential force in the current cube.

[0057] Meanwhile, since obstacles may pose a threat, a safety zone needs to be established for the detected obstacles to prevent the AUV from getting too close. Let the position of the detected obstacle be p. e =[x e ,y e ,z e The safe distance is d. s , Where R t Let D be the detection range of the AUV. Then the safe range is D. S ={p s |||p s -p e ||≤d s}, that is, expanding the obstacle to a point with its position as the center of the sphere, d s A sphere with radius . Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the underwater environment modeling method of the present invention;

[0059] Figure 2 This is a schematic diagram illustrating the division and allocation of the task area in this invention;

[0060] Figure 3 This is a schematic diagram of the line-of-sight technology of the present invention;

[0061] Figure 4 This is a schematic diagram of the method for combining global and local paths according to the present invention. Detailed Implementation

[0062] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0063] The steps of the multi-AUV cooperative method based on artificial bee colony and artificial potential field in marine search include:

[0064] Step 1: As Figure 1 As shown, the environmental modeling method for the task area is as follows. First, assume the task area is a cube, and then divide it into several smaller cubes. The side length of the smaller cubes is set according to the detection range of the AUV. For example, if the detection range of the AUV is 30-50m, then coarse-grained modeling sets the side length to 60m, so that the AUV can detect the entire cube when it is at the center of the cube, as shown in the left figure. Therefore, the center of all cubes can be taken as the target position. Fine-grained modeling sets the side length to 20m, as shown in the right figure, and uses the center of the cube as the waypoint.

[0065] Step Two: As Figure 2As shown, the task area is divided and allocated using a different method. The improved Voronoi clustering strategy replaces the time constraint in the Voronoi clustering strategy with a distance constraint between assigned and unassigned target locations, and adds a constraint on the number of assigned target locations. This improvement has two advantages: first, it avoids some AUVs being assigned too many target locations, while others are assigned too few or none, resulting in an excessively long total search time; second, it ensures that the target locations assigned to each AUV form a continuous search area.

[0066] Step 3: As Figure 3 As shown, the global path planning is performed. The artificial bee colony in the artificial bee colony algorithm is divided into three groups: hired bees, follower bees, and scout bees. For each journey segment, N paths are randomly generated, resulting in N solutions, or N nectar sources. At the beginning of the search, hired bees search around nectar source i to generate a new nectar source. The hired bees calculate the fitness values ​​of nectar source i and the new nectar source, using a greedy method to select the better nectar source. Follower bees, based on the nectar source information shared by the hired bees, select the nectar source with the highest fitness value to follow. For each nectar source, the longer the path and the larger the turning angle, the lower its fitness value. Then, the follower bees use a roulette wheel method to select hired bees, i.e., generating a uniformly distributed random number r in [0,1], if p... i If the value is greater than r, the follower bee generates a new nectar source around nectar source i in the same way as the hired bee, and uses the same greedy selection method to determine which nectar source to retain. During the search process, if nectar source i reaches the threshold l after a finite number of iterations... limit If no better nectar source is found, the nectar source will be abandoned, and the corresponding hired bee will become a scout bee. The scout bee will randomly generate a new nectar source in the search space to replace nectar source i.

[0067] Simultaneously, line-of-sight technology is used to smooth the planned global path. Line-of-sight technology, such as... Figure 3 As shown, a LOS check is performed at each pathpoint in the global path to determine whether the pathpoint can be skipped. For example... Figure 3 In the two cases shown, the gray dashed line represents the original global path, in x i,j Performing a LOS check determines whether the path can be modified to start from x. i,j The previous path point x i,j-1 Go directly to x i,j The next path point x i,j+1 Because of x i,j-1 With x i,j+1 The connection does not pass through the infeasible region, so it can be modified, and for x... i,k For example, due to x i,k-1 With x i,k+1The connection passes through an infeasible region, so it cannot be modified. For path points at the target location, no LOS check is performed, meaning these path points cannot be skipped. The modified path will be smoother and more suitable for practical AUV applications.

[0068] Step Four: As Figure 4 As shown, this illustrates the local path planning method. When the AUV detects an infeasible region on the global path, the first path point before the infeasible region is taken as the starting point of the local path, and the first path point after the infeasible region is taken as the target point of the local path. For example... Figure 4 As shown, the black dashed line represents the global path, and the black solid line represents the local path. After bypassing the infeasible region via the local path, the AUV returns to the global path. The artificial potential field in this scheme consists of the gravitational field exerted by the target point and the repulsive field exerted by the infeasible region. The magnitude of the gravitational field depends on the distance between the AUV and the target point; the greater the distance, the stronger the gravitational field. The magnitude of the repulsive field depends on both the distance between the AUV and the infeasible region and the distance between the AUV and the target point; the greater the distance between the AUV and the infeasible region, the stronger the repulsive field. Simultaneously, when the AUV reaches the target point, the magnitude of the repulsive field is 0. The corresponding gravitational and repulsive forces are the negative gradients of the gravitational and repulsive fields, respectively. The resultant force of the gravitational and repulsive forces (i.e., the potential force) will drive the AUV away from the obstacle and closer to the target point. In local path planning, the direction of the vector from the current path point to the next path point needs to be as close as possible to the direction of the potential force.

[0069] At the same time, since obstacles may pose a certain threat, it is necessary to set a safety range for the detected obstacles to prevent the AUV from getting too close, that is, to expand the outline of the obstacle into a sphere.

Claims

1. A multi-AUV cooperative method based on artificial bee colonies and artificial potential fields in marine search, characterized by: Includes the following steps: Step 1: Environmental modeling of the task area The task region is divided into feasible and infeasible regions, and coarse-grained and fine-grained cube modeling methods are used to model the environment of the task region. Step Two: Division and Allocation of Task Areas An improved Voronoi clustering strategy is used to divide the mission area into several small regions and assign them to each AUV, so that each AUV is responsible for reconnaissance of a small region. The center of the coarse-grained cube generated in step one is used as the target location of the Voronoi clustering strategy. Each target location is assigned to the AUV that takes the least time to reach it. The number of assigned target locations and continuity constraints are added to avoid uneven distribution and ensure the continuity of small regions. Step 3: Global Path Planning An artificial bee colony algorithm is used to plan a global path with a short path and a small turning angle, and a line-of-sight technique is introduced to smooth the path planned by the artificial bee colony algorithm. Step 4: Local Path Planning When an AUV navigates on the global path planned in step three, it will detect obstacles or other AUVs, and the area where they are located is the infeasible region. If the infeasible region is located on the global path, an artificial potential field method is used to plan a local path to avoid it. In step four, the local path planning method is as follows: When using the artificial potential field method for local path planning, fine-grained cube modeling is employed, with the center of each cube serving as a path point. When the AUV detects an infeasible region on the global path, the first path point in front of the infeasible region is designated as the starting point of the local path, and the first path point behind the infeasible region is designated as the target point. The artificial potential field is composed of the gravitational field exerted by the target point and the repulsive field exerted by the infeasible region. The magnitude of the gravitational field depends on the distance between the AUV and the target point; the greater the distance, the stronger the gravitational field. The magnitude of the repulsive field depends on both the distance between the AUV and the infeasible region and the distance between the AUV and the target point; the greater the distance between the AUV and the infeasible region, the stronger the repulsive field. Simultaneously, when the AUV reaches the target point, the magnitude of the repulsive field is 0. The corresponding gravitational and repulsive forces are the negative gradients of the gravitational and repulsive fields, respectively. The resultant force of the gravitational and repulsive forces will drive the AUV away from obstacles and closer to the target point. Let the current path point of the AUV be The target point is The gravitational field exerted on the AUV by the target point is designed as follows: ； in The position gain function, It is the distance between the AUV and the target point; the corresponding gravity is the negative gradient of the gravitational field function: ； During the AUV's journey to the target point, this force converges linearly to zero; the repulsive force field exerted by the infeasible region on the AUV is designed as follows: ； in The position gain function, Location of AUV The shortest distance to the obstacle. It is a constant representing the distance of the obstacle's influence; It is an arbitrary constant greater than zero. By introducing the distance between the AUV and the target, it is ensured that the entire potential field exists only at the target point. Global minimum; when When the repulsive force is expressed as: ； in, ； ； 。 2. The multi-AUV cooperative method based on artificial bee colony and artificial potential field in marine search according to claim 1, characterized in that: In step one, the environmental modeling method for the task area is as follows: First, assume the task area is a cube, then divide it into several smaller cubes. The side length of the smaller cubes is set according to the detection range of the AUV. In coarse-grained cube modeling, the center of all cubes is taken as the target position in the division and allocation of the task area. In fine-grained cube modeling, the center of the cube is taken as the path point in path planning.

3. The multi-AUV cooperative method based on artificial bee colony and artificial potential field in marine search according to claim 1, characterized in that: In step two, the method for dividing and allocating the task area is as follows: When using the improved Voronoi clustering strategy to partition and allocate task regions, a coarse-grained cube model is adopted, where the center of each cube is taken as the target location. Assuming that there are m AUVs randomly distributed outside the task region, the Voronoi clustering strategy is formulated as follows, taking into account the location of each AUV: (1) in It is the first The location of each AUV, Indicates the first An AUV arrives at the target from its location. Time required; The improved Voronoi clustering strategy changes the objective function with respect to time to an objective function with respect to the distance between assigned and unassigned target locations, and adds a constraint on the number of assigned target locations; The improved Voronoi clustering strategy is formulated as follows: （2） in It has already been assigned to the first A set of target locations for AUVs It is the set of unclustered target locations, while Indicates the target location and target location The Euclidean distance between them; It has already been assigned to the first The number of target locations for each AUV , These are two parameters that are set manually.

4. The multi-AUV cooperative method based on artificial bee colony and artificial potential field in marine search according to claim 1, characterized in that: In step three, the global path planning method is as follows: When using a combination of artificial bee colony algorithm and line-of-sight technique for global path planning, fine-grained cube modeling is adopted, in which the center of each cube is used as a path point; In the artificial bee colony algorithm, the artificial bee colony is divided into three groups: mercenary bees, follower bees, and scout bees; for each leg of the journey, the bees are randomly generated. One path, that is One solution, that is One nectar source; in the initial stage of the search, hired bees are located at the nectar source. The surrounding search generates a new nectar source; the hired bees calculate the nectar source. Based on the fitness value of new nectar sources, a greedy selection method is used to choose the better nectar source; the follower bee selects the nectar source with the highest fitness value to follow based on the nectar source information shared by the hired bee; for each nectar source, the longer the path and the larger the turning angle, the lower its fitness value; then, the follower bee uses a roulette wheel method to select the hired bee, that is, in Generate a uniformly distributed random number. If following probability Greater than The follower bee uses the same method as the hired bee to access nectar sources. A new nectar source is generated around it, and the same greedy selection method as the hired bee is used to determine which nectar source to retain; during the search process, if a nectar source... The threshold is reached after a finite number of iterations. If no better nectar source is found, the nectar source will be abandoned, and the corresponding mercenary bee will become a scout bee; the scout bee will randomly generate a new nectar source in the search space to replace the original nectar source. ; Meanwhile, line-of-sight (LOS) technology is used to smooth the planned global path: a LOS check is performed at each path point of the global path to determine whether the path point can be skipped, that is, whether the path point can be directly reached from the previous path point to the next path point; for the path point of the target location, no LOS check is performed, that is, the path point cannot be skipped.

5. The multi-AUV cooperative method based on artificial bee colony and artificial potential field in marine search according to claim 1, characterized in that: In step four, in order to stay away from obstacles, a safe range is set for the detected obstacles to prevent the AUV from getting too close, that is, the outline of the obstacle is expanded into a sphere.

Images