Method for multi-aUV task area partitioning and allocation based on clustering and resource constraints

Abstract

The present application relates to the technical field of AUV control based on swarm intelligence, and particularly relates to a multi-AUV task area division and distribution method based on clustering and resource constraints. The method comprises the following steps: target pre-screening of a task space based on a DBSCAN method; screening of low-density area targets by using a resource execution constraint screening mechanism; selection of a load placement position on the effective target set determined in step S2 by a resource range constraint mechanism, limited by the range of the load carried by the AUV; and distribution of the load placement position set obtained in step S3. The method realizes optimization of target coverage by the AUV under limited resources.

Description

Technical Field

[0001] This invention relates to the field of AUV control technology based on swarm intelligence, and in particular to a method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints. Background Technology

[0002] Advances in marine robotics technology have significantly improved the intelligence level of AUVs (Autonomous Marine Vehicles). However, single AUVs, limited by coverage, payload, and fault tolerance, struggle to handle complex marine missions. Multi-AUV collaborative operations, through parallel execution and functional complementarity, can achieve large-scale, efficient exploration and data acquisition, significantly improving system performance. Mission area division and allocation are core technologies for multi-AUV collaborative operations. Existing research primarily focuses on full-coverage missions under ideal conditions, generally neglecting resource constraints such as AUV endurance and payload, and has not yet developed a systematic solution for resource-constrained scenarios. Conducting research on mission area division and allocation for multi-AUVs can not only improve the execution efficiency of tasks such as seabed ecological monitoring and resource exploration, but its algorithmic framework can also be extended to terrestrial scenarios such as wireless base station site selection and sensor network deployment, possessing significant theoretical and applied value.

[0003] The Set Cover Problem (SCP) aims to select the minimum number of sets to cover all demand points. This application addresses the partitioning and allocation of AUV task areas under resource constraints, specifically selecting sets to cover as many demand points as possible and allocate them to AUVs—the Max Coverage Problem (MCP).

[0004] Research on maximum cover algorithms has evolved from geometric segmentation to heuristic optimization and then to intelligent learning. Geometric segmentation-based region partitioning relies on geometric coverage and rule-driven strategies, such as grid partitioning and sector coverage, and constructs mathematical models to obtain optimal solutions for task allocation. Methods based on geometric segmentation and mathematical modeling for task allocation offer advantages such as accurate solutions and controllable constraints, but their computational complexity increases rapidly with the scale of the task.

[0005] With the increasing scale of tasks and the growing demand for intelligence, heuristic methods have been applied to region partitioning, leading to the development of cluster-based coverage optimization methods such as K-means. These methods extract the spatial structure of task points to partition and allocate task regions. While this improves overall coverage efficiency and execution feasibility, such methods struggle to simultaneously consider differences in task value, path costs, and energy consumption balance. Furthermore, they are prone to issues such as regional connectivity problems or uneven load distribution in complex terrains.

[0006] Furthermore, based on heuristic ideas, researchers have begun to extend their research towards multi-objective optimization, introducing optimization algorithms such as genetic algorithms and particle swarm optimization. This method jointly optimizes maximum coverage with path length, energy consumption constraints, and other factors. While these algorithms can obtain coverage solutions closer to the global optimum, their computational cost increases significantly with the task size, making them unsuitable for scenarios with changing real-time coverage or online task requirements.

[0007] In recent years, reinforcement learning algorithms have been widely used, and this method has been introduced into the maximum coverage problem. It achieves adaptive optimization of region selection by learning the strategic relationship between task point distribution, coverage gains, and energy consumption. However, reinforcement learning has high training costs, strong dependence on large amounts of data, and high requirements for airborne computing power and underwater acoustic communication bandwidth, which limits its application in engineering scenarios. Currently, heuristic algorithms remain the mainstream in practical applications. Therefore, this application proposes a multi-AUV task region partitioning and allocation method based on heuristic algorithms, combining the value function optimization ideas of clustering algorithms and intelligent swarm algorithms. Summary of the Invention

[0008] The purpose of this invention is to overcome the above-mentioned defects in the existing technology and propose a multi-AUV task area partitioning and allocation method based on clustering and resource constraints, which realizes the optimization of AUV target coverage under limited resources.

[0009] The technical solution of this invention is: a method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints, comprising the following steps:

[0010] S1. Pre-screening of targets in the task space based on the DBSCAN method;

[0011] S2. Use the resource execution constraint screening mechanism to screen targets in low-density areas;

[0012] S3. Limited by the effective range of the payload carried by the AUV, the payload placement location is selected on the effective target set determined in step S2 through the resource range constraint mechanism.

[0013] S4. Allocate the load placement location set obtained in step S3.

[0014] In this invention, in step S1,

[0015] Each target location The neighborhood of is:

[0016] ,

[0017] in, Indicates the radius of the domain; and Representing the target and The location, and ; Let the set of target locations be defined, and let:

[0018] ,

[0019] in, Indicates the number of targets; The x-coordinate value representing the target location; The vertical coordinate value representing the target location;

[0020] Based on the neighborhood information of each target, a core target set after target filtering is obtained. :

[0021] ,

[0022] in, Indicates the minimum number of targets in the neighborhood;

[0023] Starting from each core target point, recursively aggregate target points that are density-reachable from it to form several target density sub-clusters:

[0024] ,

[0025] in, Indicates the number of target density subclusters; Indicates the first A target density sub-cluster;

[0026] For target points that do not belong to any target density subcluster, a low-density target set is formed:

[0027] ,

[0028] in, This represents a low-density target set.

[0029] Let the first AUV in the current iteration round The current position is The remaining navigation resources are The remaining payload resources are ;

[0030] Suppose the AUV travels from its current location to the target point. After completing the mission, the player can return to the recovery point or continue with subsequent missions. The minimum travel cost is defined as follows:

[0031] ,

[0032] in, Indicates the first AUV and the first The distance between the target points;

[0033] If the target point The required task payload consumption is The target point is considered to be true when the following conditions are met. For the first AUV resources are available:

[0034] ,

[0035] and

[0036] ,

[0037] To avoid path spikes caused by adding a target point to a task cluster, we define a method to add the target point to the first task cluster. The path increment after each task cluster is:

[0038] ,

[0039] in, Indicates the first The path length of the AUV in the current iteration round corresponding to the task cluster; Indicates the target point The new path length after joining the task cluster;

[0040] When satisfied At that time, the target point is considered This will not cause the corresponding task cluster path to exceed the limit. Indicates the path increment threshold;

[0041] Let the resource reachability determination function be:

[0042] ,

[0043] If there exists at least one AUV that satisfies If the target point is not specified, then retain that target point; if it is specified for all AUVs... If so, the target point is determined to be an unreachable point.

[0044] The effective target set is composed of low-density target points that satisfy resource accessibility and target points in each target density sub-cluster.

[0045] ,

[0046] Low-density target points that do not meet the conditions are defined as the candidate target set.

[0047] ,

[0048] The above candidate target set will be discarded.

[0049] In step S3,

[0050] Based on the effective target set Calculate the set of candidate load placement locations:

[0051] ,

[0052] in, Indicates the first The coordinates of the candidate load placement locations; Indicates the number of candidate payload placement locations;

[0053] The resource range constraint mechanism uses a cyclic iterative approach to select the optimal load placement location, choosing any candidate load placement location. As the initial value, its coverable target subset is defined as:

[0054] ,

[0055] in, Indicates the effective coverage radius of the AUV payload;

[0056] Let the first The set of targets that have not yet been covered during the round of iterations is:

[0057] ,

[0058] The initial value is:

[0059] ,

[0060] Then the candidate load placement position In the The new set of coverage targets in each iteration is:

[0061] ,

[0062] The number of newly added coverage targets is:

[0063] ,

[0064] For candidate load placement locations Let it be assigned to the first The cost of sailing with an AUV is:

[0065] ,

[0066] in, Indicates the first AUV and payload candidate deployment locations The distance between them; The weighting coefficient representing the cost of navigation can be adjusted according to the degree of importance attached to the cost of navigation. Generally, ;

[0067] Let the load consumption required to execute the load candidate placement position be... The path increment caused by adding it is:

[0068] ,

[0069] in, Indicates the candidate placement location of the payload. The new path length after joining the task cluster;

[0070] If both conditions are met:

[0071] ,

[0072] ,

[0073] ,

[0074] Then the candidate load placement position is considered For the first The AUV is executable;

[0075] Definition of the first The set of candidate load placement locations for the wheel is as follows:

[0076] ,

[0077] The candidate load placement locations that do not meet the conditions constitute the frozen candidate set:

[0078] ,

[0079] For the first Any load candidate placement position in the set of executable load candidate placement positions for the wheel The marginal effective coverage benefit is defined as:

[0080] ,

[0081] in, Weight parameters representing path increments; Weighting parameters representing the number of loads; Indicates the candidate load placement position The AUV number that satisfies resource executable constraints and has the lowest overall cost;

[0082] No. The optimal load distribution position of the wheel is defined as:

[0083] ,

[0084] The corresponding AUV number is:

[0085] ,

[0086] And the The above resource executable constraints must be met;

[0087] Complete the first After wheel selection, update the set of optimal load placement locations:

[0088] ,

[0089] in, Indicates the current The set of optimal load placement locations at each moment, initially ;

[0090] Update the set of uncovered targets:

[0091] ,

[0092] in, Indicates the current The set of targets already covered at any given time;

[0093] Update the remaining navigation resources for the corresponding AUV:

[0094] ,

[0095] in, Indicates the current Remaining navigation resources at any given time; Indicates the current The navigation resources consumed in determining the deployment location to be reached at any given moment;

[0096] Update the remaining payload resources for the corresponding AUV:

[0097] ,

[0098] in, Indicates the current Remaining payload resources at any given time; Indicates the current The payload resources consumed at the deployment location that will be deployed at any time;

[0099] And update its current location:

[0100] ,

[0101] After each round of state updates, the frozen candidate set is... Each candidate placement location is re-evaluated for resource executableness; if a candidate placement location meets the execution conditions under the new resource state, it is moved back into the executable candidate set.

[0102] This yields the set of optimal load placement locations:

[0103] ,

[0104] in, This represents the total number of iterations at the time of termination.

[0105] In step S3, the iterative process is repeated until any of the following termination conditions are met:

[0106] (1) The set of uncovered targets is empty;

[0107] (2) The set of executable payload candidate placement locations is empty;

[0108] (3) The remaining payload of all AUVs is insufficient to support the new mission;

[0109] (4) All remaining AUV navigation resources are insufficient to support new missions.

[0110] The specific implementation process of step S4 is as follows:

[0111] S4.1. Use K-means clustering to construct the initial task clusters and assign AUVs;

[0112] S4.2, For any load placement position and any of the following is marked as For AUVs, construct a joint cost function;

[0113] S4.3 Utilize a weighted cluster center update strategy to shift the cluster center towards the core deployment location where the unit cost covers higher benefits;

[0114] S4.4 When path imbalance and task load imbalance occur simultaneously, the boundary task point with the minimum original cluster coverage loss and the minimum target cluster path increment is migrated through the resource balancing constraint mechanism.

[0115] In step S4.1,

[0116] Using the optimal load placement location set as input, the initial task cluster is constructed and AUVs are assigned according to the joint cost of distance, path increment, load occupancy and local density.

[0117] From the set of optimal load placement locations Selected from 1 location as the initial center set:

[0118] ,

[0119] in, Indicates the first An initial center.

[0120] In step S4.2,

[0121] Based on the actual mission requirements of AUVs, the first... The joint cost function for round iterations is:

[0122] ,

[0123] in, This indicates the distance between the load placement location and the current cluster center; This indicates the location of the load placement and is included in the first... Path increment after each task cluster; This indicates the increase in AUV payload consumption after incorporation; Indicates the load placement location Local density or coverage value of surrounding targets; These are the weighting coefficients;

[0124] When the load is placed For the first When an AUV does not satisfy the remaining resource constraint, let:

[0125] ,

[0126] This prevents non-executable ownership relationships from participating in the allocation;

[0127] Place each load at a designated location. Assign the AUV with the lowest joint cost:

[0128] ,

[0129] Therefore, we obtain the first... The set of task clusters in round iteration:

[0130] ,

[0131] in, Indicates assignment to the first A cluster of AUV tasks.

[0132] In step S4.3,

[0133] Based on the benefits and costs of each payload placement location in the current task cluster, update the cluster center to a weighted center:

[0134] ,

[0135] Among them, weight Defined as:

[0136] ,

[0137] When satisfied At that time, it is assumed that the initial task cluster construction is convergent. Indicates the central convergence threshold;

[0138] This yields the initial set of task clusters:

[0139] .

[0140] In step S4.4,

[0141] Let the first AUV corresponding to task cluster The total length of the navigation path is:

[0142] ,

[0143] in, Indicates the first AUV and payload candidate deployment locations The distance between them; Represents task cluster The first in the order of execution One placement location;

[0144] Define the path ratio as:

[0145] ,

[0146] in, Used to measure the dispersion of each AUV mission path;

[0147] Definition of the first The mission payload of each AUV is:

[0148] ,

[0149] Define the load occupancy rate as:

[0150] ,

[0151] Let the path balance threshold be The load balance threshold is When satisfied ,and At that time, the task cluster adjustment is triggered;

[0152] Let the current overloaded cluster be:

[0153] ,

[0154] in, Weighting coefficients representing path length; The weighting coefficient representing the load occupancy rate, and ;

[0155] In overload clusters In the process, boundary task points are selected as candidate migration points, and for any candidate migration point... The migration cost is defined as:

[0156] ,

[0157] in, Indicates candidate migration points Migration to receiver cluster The resulting path increment; This represents the increase in the task payload of the AUV corresponding to the receiving cluster after migration; Indicates candidate migration points Coverage loss or cluster structure disturbance caused by removal from the original cluster; , , These are the weighting coefficients;

[0158] satisfy ,and receiver cluster Comparisons are allowed, among which Indicates the load consumption of candidate migration points; Indicates receiving cluster The total length of the navigation route; Indicates the maximum allowed total voyage path length;

[0159] Choose the migration plan that minimizes the migration cost:

[0160] ,

[0161] Candidate migration points Remove from the original cluster and reassign to the target cluster. :

[0162] ,

[0163] in, Indicates candidate migration points The new overloaded cluster after being removed from the original cluster; Indicates an overloaded cluster;

[0164] ,

[0165] in, This indicates adding candidate migration points to the original target cluster. The new target cluster;

[0166] After the update, the relevant cluster centers were recalculated:

[0167] ,

[0168] And recalculate the path length and task load for each cluster;

[0169] When satisfied and At this point, the dynamic balancing adjustment ends, resulting in the final set of task clusters:

[0170] ,

[0171] The final output includes the task cluster for each AUV and its corresponding deployment location set:

[0172] ,

[0173] in, Indicates the first The AUV ultimately executes a cluster of tasks.

[0174] The beneficial effects of this invention are:

[0175] (1) This application takes "DBSCAN + improved K-means clustering algorithm" as the core framework, integrates the global optimization idea of ​​cost function of intelligent swarm algorithm, and solves the problem of multi-AUV task area division and allocation under multiple constraints path cost, load resources, coverage benefits and load balancing.

[0176] (2) This application proposes a region partitioning algorithm based on DBSCAN. By designing a resource execution constraint screening mechanism and a resource range constraint mechanism, it can optimize the target coverage of AUV under limited resources when robot resources are limited. It performs validity preprocessing on the target to reduce the interference of low-value and resource-unexecutable targets on the subsequent task cluster construction, thus achieving the optimization of target coverage of AUV under limited resources. It also takes into account both target coverage and resource constraints when selecting targets.

[0177] (3) To address the problem that traditional K-means clustering only relies on the geometric distance between sample points and cluster centers for task allocation, making it difficult to simultaneously consider path cost, resource consumption, and local coverage value, a joint cost function is constructed. This function incorporates cluster center distance, path increment, load consumption increment, and local coverage value into the task allocation process, thereby improving the quality of the initial task cluster construction. Global optimization is performed on the multi-dimensional and multi-constraint objective function, which can adaptively optimize the weight allocation of each dimension in the joint cost function and find the optimal sample allocation rule that takes into account multiple constraints.

[0178] (4) The weighted cluster center update mechanism proposed in this application mainly breaks the limitation of traditional K-means using only geometric distance to update cluster centers, and optimizes cluster centers by comprehensively considering the target benefit and cost factors; through the iterative optimization of particle swarm, the optimal weighting coefficients of benefit and cost are adaptively solved, so that the cluster center update results simultaneously meet the dual requirements of maximizing coverage benefit and minimizing resource cost, thereby improving the consistency between clustering results and actual task requirements.

[0179] (5) In response to the problems of uneven task cluster load, excessive path length difference and local cluster overload after K-means clustering, this invention proposes a resource balance constraint mechanism. Under the premise of satisfying the resource limit of a single AUV, this mechanism can achieve a dual balance of path length and load of each AUV, effectively avoid the problem of resource imbalance, and improve the execution efficiency and energy consumption balance of the AUV group system. Attached Figure Description

[0180] Figure 1 This is a flowchart of the method described in this application;

[0181] Figure 2 This is a graph showing the impact of the minimum number of neighborhood targets on the global allocation;

[0182] Figure 3 This is the result of the previous multi-AUV task area division and allocation;

[0183] Figure 4 The result of multi-AUV task region division and allocation obtained using the method described in this application;

[0184] Figure 5 This is a comparison chart of the path lengths of each AUV before and after the improvement;

[0185] Figure 6 This is a comparison chart of the number of loads for each AUV before and after the improvement. Detailed Implementation

[0186] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0187] Specific details are set forth in the following description to provide a full understanding of the invention. However, the invention can be practiced in many ways other than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0188] The multi-AUV task region partitioning and allocation method based on clustering and resource constraints described in this application includes the following steps, the flowchart of which is shown below. Figure 1 As shown.

[0189] The first step is to perform target pre-screening in the task space based on the DBSCAN method, and to perform overall structured screening of targets in the task region to reduce the screening space under subsequent resource constraints.

[0190] Based on the low-density and high-density targets output by traditional DBSCAN, this paper further preprocesses the targets by designing resource execution constraint screening mechanisms and resource range constraint mechanisms. This reduces the interference of low-value and resource-inoperable targets on the construction of subsequent task clusters, thereby optimizing the target coverage of AUVs under limited resources.

[0191] Define each target location The neighborhood of is:

[0192] ,

[0193] in, Indicates the radius of the domain; and Representing the target and The location, and ; Let the set of target locations be defined, and let:

[0194] ,

[0195] in, Indicates the number of targets; The x-coordinate value representing the target location; The vertical coordinate value represents the target location.

[0196] Based on the neighborhood information of each target, a core target set after target filtering is obtained. :

[0197] ,

[0198] in, This represents the minimum number of targets in the domain.

[0199] Starting from each core target point, recursively aggregate target points that are density-reachable from it to form several target density subclusters:

[0200] ,

[0201] in, Indicates the number of target density subclusters; Indicates the first A target density sub-cluster.

[0202] For target points that do not belong to any target density subcluster, a low-density target set is formed:

[0203] ,

[0204] in, This represents a low-density target set.

[0205] The second step utilizes a resource-constrained screening mechanism to filter targets in low-density areas, achieving a dual balance between the path length and payload of each AUV. This ensures full utilization of AUV resources to maximize area coverage. This effectively avoids resource imbalance issues and improves the execution efficiency and energy consumption balance of the AUV swarm system.

[0206] For any low-density target point obtained in the first step Instead of deleting it directly, it determines whether it is executable for each AUV.

[0207] Let the first AUV in the current iteration round The current position is The remaining navigation resources are The remaining payload resources are If the AUV reaches the target point from its current location. After completing the mission, the player can return to the recovery point or continue with subsequent missions. The minimum travel cost is defined as follows:

[0208] ,

[0209] in, Indicates the first AUV and the first The distance between the target points.

[0210] If the target point The required task payload consumption is The target point is considered to be true when the following conditions are met. For the first AUV resources are available:

[0211] ,

[0212] and

[0213] ,

[0214] Furthermore, to avoid a sudden increase in path length after adding a target point to a task cluster, a definition is also defined for its addition to the first task cluster. The path increment after each task cluster is:

[0215] ,

[0216] in, Indicates the first The path length of the AUV in the current iteration round corresponding to the task cluster; Indicates the target point The new path length after joining the task cluster.

[0217] When satisfied At that time, the target point is considered This will not cause the corresponding task cluster path to exceed the limit. This represents the path increment threshold.

[0218] In summary, the resource reachability determination function is defined as follows:

[0219] ,

[0220] If there exists at least one AUV that satisfies If the target point is not specified, then retain that target point; if it is specified for all AUVs... If so, the target point is determined to be an unreachable point.

[0221] The effective target set is composed of low-density target points that satisfy resource accessibility and target points in each target density sub-cluster.

[0222] ,

[0223] Low-density target points that do not meet the conditions are defined as the candidate target set.

[0224] ,

[0225] The above candidate target set will be discarded.

[0226] The third step involves selecting a suitable AUV arrival point, i.e. the deployment location of the payload, from the effective target set determined in the second step, based on the limited range of the payload carried by the AUV, such as the monitoring device.

[0227] Based on the effective target set Calculate the set of candidate load placement locations:

[0228] ,

[0229] in, Indicates the first The coordinates of the candidate load placement locations; This indicates the number of candidate payload placement locations.

[0230] The resource range constraint mechanism uses a cyclic iterative approach to select the optimal load placement location, choosing any candidate load placement location. Let the initial value be , and define the target subset that it can cover as:

[0231] ,

[0232] in, Indicates the effective coverage radius of the AUV payload.

[0233] Let the first The set of targets that have not yet been covered during the round of iterations is:

[0234] ,

[0235] The initial value is:

[0236] ,

[0237] Then the candidate load placement position In the The new set of coverage targets in each iteration is:

[0238] ,

[0239] The number of newly added coverage targets is:

[0240] ,

[0241] For candidate load placement locations Let it be assigned to the first The cost of sailing with an AUV is:

[0242] ,

[0243] in, Indicates the first AUV and payload candidate deployment locations The distance between them; The weighting coefficient representing the cost of navigation can be adjusted according to the degree of importance attached to the cost of navigation. Generally, .

[0244] Let the load consumption required to execute the load candidate placement position be... The path increment caused by adding it is:

[0245] ,

[0246] in, Indicates the candidate placement location of the payload. The new path length after joining the task cluster.

[0247] If both conditions are met:

[0248] ,

[0249] ,

[0250] ,

[0251] Then the candidate load placement position is considered For the first Each AUV is executable. Therefore, the first... The set of candidate load placement locations for the wheel is as follows:

[0252] ,

[0253] The candidate load placement locations that do not meet the conditions constitute the frozen candidate set:

[0254] ,

[0255] For the first Any load candidate placement position in the set of executable load candidate placement positions for the wheel To ensure that, under the constraints of limited AUV navigation resources and payload capacity, the benefits of new coverage are balanced with the overall execution cost, and to avoid the drawback of pure coverage-first algorithms prematurely exhausting resources and causing a decline in global coverage performance, the marginal effective coverage benefit is defined as follows:

[0256] ,

[0257] in, Weight parameters representing path increments; Weighting parameters representing the number of loads, and The AUV's resource limitations can be flexibly adjusted based on whether the priority is extended range or payload. Indicates the candidate load placement position The AUV number that satisfies resource executable constraints and has the lowest overall cost.

[0258] No. The optimal load distribution position of the wheel is defined as:

[0259] ,

[0260] The corresponding AUV number is:

[0261] ,

[0262] And the The above resource executable constraints are satisfied.

[0263] Complete the first After wheel selection, update the set of optimal load placement locations:

[0264] ,

[0265] in, Indicates the current The set of optimal load placement locations at each moment, initially .

[0266] Update the set of uncovered targets:

[0267] ,

[0268] in, Indicates the current The set of targets already covered at any given time;

[0269] Update the remaining navigation resources for the corresponding AUV:

[0270] ,

[0271] in, Indicates the current Remaining navigation resources at any given time; Indicates the current The navigation resources consumed at the deployment location to be reached at any time.

[0272] Update the remaining payload resources for the corresponding AUV:

[0273] ,

[0274] in, Indicates the current Remaining payload resources at any given time; Indicates the current The payload resources consumed at the deployment location that will be deployed at any time.

[0275] And update its current location:

[0276] ,

[0277] After each round of state updates, the frozen candidate set is... Each candidate placement location is re-evaluated for resource executableness. If a candidate placement location meets the execution conditions under the new resource state, it is moved back into the executable candidate set.

[0278] Repeat the above iterative process until any of the following termination conditions are met: (1) the set of uncovered targets is empty; (2) the set of executable payload candidate deployment locations is empty; (3) all remaining AUV payloads are insufficient to support new missions; (4) all remaining AUV navigation resources are insufficient to support new missions.

[0279] This yields the set of optimal load placement locations:

[0280] ,

[0281] in, This represents the total number of iterations at the time of termination.

[0282] The fourth step is to allocate the set of optimal load placement locations determined in the third step.

[0283] The optimal payload deployment location set represents the mission execution locations of the AUVs, such as the best observation location, the best strike location, and the best mapping location. Since multiple AUVs are used to complete the mission, this step requires allocating the optimal payload deployment locations to ensure that the mission load and flight path of each AUV are as balanced as possible, thereby avoiding any AUV from being overworked or having an excessively long flight path.

[0284] First, K-means clustering is used to construct the initial task clusters and assign AUVs.

[0285] Unlike the target area division mentioned above, this part is the full coverage task allocation for a fixed number of AUVs. Since the K-means clustering algorithm can generate corresponding task clusters based on the number of AUVs, this application is based on K-means clustering.

[0286] Using the set of optimal payload placement locations as input, the initial task clusters and AUV assignments are constructed based on the joint costs of distance, path increment, payload occupancy, and local density, and are also implemented in an iterative manner.

[0287] From the set of optimal load placement locations Selected from 1 location as the initial center set:

[0288] ,

[0289] in, Indicates the first An initial center.

[0290] Second, for any load placement location and any of the following is marked as AUVs are used to construct a joint cost function.

[0291] To address the issue that traditional K-means clustering relies solely on spatial Euclidean distance, a new method is defined based on the actual task requirements of AUVs. The joint cost function for round iterations is:

[0292] ,

[0293] in, This indicates the distance between the load placement location and the current cluster center; This indicates the location of the load placement and is included in the first... Path increment after each task cluster; This indicates the increase in AUV payload consumption after incorporation; Indicates the load placement location Local density or coverage value of surrounding targets; This is a weighting coefficient, which can be determined based on mission priority and the actual endurance and payload cost of the AUV.

[0294] When the load is placed For the first When an AUV does not satisfy the remaining resource constraint, let:

[0295] ,

[0296] This prevents non-executable ownership relationships from participating in the allocation.

[0297] Place each load at a designated location. Assign the AUV with the lowest joint cost:

[0298] ,

[0299] Therefore, we obtain the first... The set of task clusters in round iteration:

[0300] ,

[0301] in, Indicates assignment to the first A cluster of AUV tasks.

[0302] Third, to address the problem of indiscriminate clustering in traditional K-means based solely on spatial geometric mean, this application proposes a weighted cluster center update strategy. This strategy shifts the cluster centers towards core deployment locations with higher coverage per unit cost, resulting in more compact task clusters that balance task value with AUV resource consumption.

[0303] Based on the benefits and costs of each payload placement location in the current task cluster, update the cluster center to a weighted center:

[0304] ,

[0305] Among them, weight Defined as:

[0306] ,

[0307] in, This indicates the increase in AUV load consumption after incorporation.

[0308] When satisfied At that time, the initial task cluster construction is considered convergent. Among them, This represents the central convergence threshold.

[0309] This yields the initial set of task clusters:

[0310] .

[0311] Fourth, when path imbalance and task load imbalance occur simultaneously, the boundary task point that minimizes the original cluster coverage loss and the target cluster path increment is migrated through the resource balancing constraint mechanism.

[0312] Let the first AUV corresponding to task cluster The total length of the navigation path is:

[0313] ,

[0314] in, Indicates the first AUV and payload candidate deployment locations The distance between them; Represents task cluster The first in the order of execution Placement locations.

[0315] Define the path ratio as:

[0316] ,

[0317] in, Used to measure the dispersion of each AUV mission path.

[0318] Definition of the first The mission payload of each AUV is:

[0319] ,

[0320] Define the load occupancy rate as:

[0321] ,

[0322] Let the path balance threshold be The load balance threshold is Task cluster adjustments are triggered when the following conditions are met:

[0323] ,

[0324] and

[0325] ,

[0326] That is, subsequent adjustments will only be made when the path differences are too large and at least one AUV task has an excessively high load.

[0327] Let the current overloaded cluster be:

[0328] ,

[0329] in, Weighting coefficients representing path length; The weighting coefficient representing the load occupancy rate, and If path length is important, then ;on the contrary, If equal importance is attached, then .

[0330] In overload clusters In this process, boundary task points are selected as candidate migration points. To address the load imbalance issue caused by the initial K-means clustering, which results in some AUVs being severely overloaded while others are idle, and to ensure dynamic load balancing among AUVs while minimizing sacrifices in global coverage performance and increases in total resource consumption, for any candidate migration point... The migration cost is defined as:

[0331] ,

[0332] in, Indicates candidate migration points Migration to receiver cluster The resulting path increment; This represents the increase in the task payload of the AUV corresponding to the receiving cluster after migration; Indicates candidate migration points Coverage loss or cluster structure disturbance caused by removal from the original cluster; , , This is a weighting coefficient, which can be determined based on the actual range, payload cost, and coverage loss of the AUV.

[0333] satisfy ,and receiver cluster Only those who are allowed to participate in the comparison, among which Indicates the load consumption of candidate migration points; Indicates receiving cluster The total length of the navigation route; .

[0334] Choose the migration plan that minimizes the migration cost:

[0335] ,

[0336] Candidate migration points Remove from the original cluster and reassign to the target cluster. :

[0337] ,

[0338] in, Indicates candidate migration points The new overloaded cluster after being removed from the original cluster; Indicates an overloaded cluster;

[0339] ,

[0340] in, This indicates adding candidate migration points to the original target cluster. The new target cluster.

[0341] After the update, the relevant cluster centers were recalculated:

[0342] ,

[0343] And recalculate the path length and task load for each cluster.

[0344] When satisfied and At this point, the dynamic balancing adjustment ends, resulting in the final set of task clusters:

[0345] ,

[0346] The final output includes the task cluster for each AUV and its corresponding deployment location set:

[0347] ,

[0348] in, Indicates the first The AUV ultimately executes a cluster of tasks.

[0349] To verify the effectiveness of the method proposed in this application, simulation verification was performed on the multi-AUV task region partitioning and allocation method proposed in this application. The experiment set up four AUVs, and the parameters in the method were specifically set as follows: , , , , , , , , , .

[0350] The partitioning of regions is related to the minimum number of neighborhood targets selected for each cluster center, and its impact on the overall allocation is shown in the figure. Therefore, the target number chosen should balance coverage and resource constraints, i.e., take... And randomly set target locations in space, the final allocation result is as follows As shown. With Compared to the previous allocation shown, the method proposed in this application balances the task load of each AUV and does not exceed its maximum load. Furthermore, the path lengths of each AUV before and after the improvement are compared as follows: As shown, the number of payloads for each AUV before and after the improvement is compared. As shown, before the improvement, the path length and payload of each AUV varied greatly, while after the improvement, the path length and payload of each AUV are more balanced. In summary, the method proposed in this application can realize the division and balanced allocation of task areas for multiple AUVs under limited resources.

[0351] The above provides a detailed description of the multi-AUV task region partitioning and allocation method based on clustering and resource constraints provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention. The above description of the disclosed embodiments enables those skilled in the art to implement or use this invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of this invention. Therefore, this invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints, characterized in that, Includes the following steps: S1. Pre-screening of targets in the task space based on the DBSCAN method; S2. Use the resource execution constraint screening mechanism to screen targets in low-density areas; S3. Limited by the effective range of the payload carried by the AUV, the payload placement location is selected on the effective target set determined in step S2 through the resource range constraint mechanism. S4. Allocate the load placement location set obtained in step S3.

2. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 1, characterized in that, In step S1, Each target location The neighborhood of is: ， in, Indicates the radius of the domain; and Representing the target and The location, and ; Let the set of target locations be defined, and let: ， in, Indicates the number of targets; The x-coordinate value representing the target location; The vertical coordinate value representing the target location; Based on the neighborhood information of each target, a core target set after target filtering is obtained. : ， in, Indicates the minimum number of targets in the neighborhood; Starting from each core target point, recursively aggregate target points that are density-reachable from it to form several target density sub-clusters: ， in, Indicates the number of target density subclusters; Indicates the first A target density sub-cluster; For target points that do not belong to any target density subcluster, a low-density target set is formed: ， in, This represents a low-density target set.

3. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 1, characterized in that, Let the first AUV in the current iteration round The current position is The remaining navigation resources are The remaining payload resources are ; Suppose the AUV travels from its current location to the target point. After completing the mission, the player can return to the recovery point or continue with subsequent missions. The minimum travel cost is defined as follows: ， in, Indicates the first AUV and the first The distance between the target points; If the target point The required task payload consumption is The target point is considered to be true when the following conditions are met. For the first AUV resources are available: ， and ， To avoid path spikes caused by adding a target point to a task cluster, we define a method to add the target point to the first task cluster. The path increment after each task cluster is: ， in, Indicates the first The path length of the AUV in the current iteration round corresponding to the task cluster; Indicates the target point The new path length after joining the task cluster; When satisfied At that time, the target point is considered This will not cause the corresponding task cluster path to exceed the limit. Indicates the path increment threshold; Let the resource reachability determination function be: ， If there exists at least one AUV that satisfies If the target point is not specified, then retain that target point; if it is specified for all AUVs... If so, the target point is determined to be an unreachable point. The effective target set is composed of low-density target points that satisfy resource accessibility and target points in each target density sub-cluster. ， Low-density target points that do not meet the conditions are defined as the candidate target set. ， The above candidate target set will be discarded.

4. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 1, characterized in that, In step S3, Based on the effective target set Calculate the set of candidate load placement locations: ， in, Indicates the first The coordinates of the candidate load placement locations; Indicates the number of candidate payload placement locations; The resource range constraint mechanism uses a cyclic iterative approach to select the optimal load placement location, choosing any candidate load placement location. As the initial value, its coverable target subset is defined as: ， in, Indicates the effective coverage radius of the AUV payload; Let the first The set of targets that have not yet been covered during the round of iterations is: ， The initial value is: ， Then the candidate load placement position In the The new set of coverage targets in each iteration is: ， The number of newly added coverage targets is: ， For candidate load placement locations Let it be assigned to the first The cost of sailing with an AUV is: ， in, Indicates the first AUV and payload candidate deployment locations The distance between them; Weighting coefficients representing the cost of navigation; Let the load consumption required to execute the load candidate placement position be... The path increment caused by adding it is: ， in, Indicates the candidate placement location of the payload. The new path length after joining the task cluster; If both conditions are met: ， ， ， Then the candidate load placement position is considered For the first The AUV is executable; Definition of the first The set of candidate load placement locations for the wheel is as follows: ， The candidate load placement locations that do not meet the conditions constitute the frozen candidate set: ， For the first Any load candidate placement position in the set of executable load candidate placement positions for the wheel The marginal effective coverage benefit is defined as: ， in, Weight parameters representing path increments; Weighting parameters representing the number of loads; Indicates the candidate load placement position The AUV number that satisfies resource executable constraints and has the lowest overall cost; No. The optimal load distribution position of the wheel is defined as: ， The corresponding AUV number is: ， And the The above resource executable constraints must be met; Complete the first After wheel selection, update the set of optimal load placement locations: ， in, Indicates the current The set of optimal load placement locations at each moment, initially ; Update the set of uncovered targets: ， in, Indicates the current The set of targets already covered at any given time; Update the remaining navigation resources for the corresponding AUV: ， in, Indicates the current Remaining navigation resources at any given time; Indicates the current The navigation resources consumed in determining the deployment location to be reached at any given moment; Update the remaining payload resources for the corresponding AUV: ， in, Indicates the current Remaining payload resources at any given time; Indicates the current The payload resources consumed at the deployment location that will be deployed at any time; And update its current location: ， After each round of state updates, the frozen candidate set is... Each candidate placement location is re-evaluated for resource executableness; if a candidate placement location meets the execution conditions under the new resource state, it is moved back into the executable candidate set. This yields the set of optimal load placement locations: ， in, This represents the total number of iterations at the time of termination.

5. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 4, characterized in that, In step S3, the iterative process is repeated until any of the following termination conditions are met: (1) The set of uncovered targets is empty; (2) The set of executable payload candidate placement locations is empty; (3) The remaining payload of all AUVs is insufficient to support the new mission; (4) All remaining AUV navigation resources are insufficient to support new missions.

6. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 1, characterized in that, The specific implementation process of step S4 is as follows: S4.

1. Use K-means clustering to construct the initial task clusters and assign AUVs; S4.2, For any load placement position and any of the following is marked as For AUVs, construct a joint cost function; S4.3 Utilize a weighted cluster center update strategy to shift the cluster center towards the core deployment location where the unit cost covers higher benefits; S4.4 When path imbalance and task load imbalance occur simultaneously, the boundary task point with the minimum original cluster coverage loss and the minimum target cluster path increment is migrated through the resource balancing constraint mechanism.

7. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 6, characterized in that, In step S4.1, Using the optimal load placement location set as input, the initial task cluster is constructed and AUVs are assigned according to the joint cost of distance, path increment, load occupancy and local density. From the set of optimal load placement locations Selected from 1 location as the initial center set: ， in, Indicates the first An initial center.

8. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 6, characterized in that, In step S4.2, Based on the actual mission requirements of AUVs, the first... The joint cost function for round iterations is: ， in, This indicates the distance between the load placement location and the current cluster center; This indicates the location of the load placement and is included in the first... Path increment after each task cluster; This indicates the increase in AUV payload consumption after incorporation; Indicates the load placement location Local density or coverage value of surrounding targets; These are the weighting coefficients; When the load is placed For the first When an AUV does not satisfy the remaining resource constraint, let: ， This prevents non-executable ownership relationships from participating in the allocation; Place each load at a designated location. Assign the AUV with the lowest joint cost: ， Therefore, we obtain the first... The set of task clusters in round iteration: ， in, Indicates assignment to the first A cluster of AUV tasks.

9. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 6, characterized in that, In step S4.3, Based on the benefits and costs of each payload placement location in the current task cluster, update the cluster center to a weighted center: ， Among them, weight Defined as: ， When satisfied At that time, it is assumed that the initial task cluster construction is convergent. Indicates the central convergence threshold; This yields the initial set of task clusters: 。 10. The method for partitioning and allocating multi-AUV task regions based on clustering and resource constraints according to claim 6, characterized in that, In step S4.4, Let the first AUV corresponding to task cluster The total length of the navigation path is: ， in, Indicates the first AUV and payload candidate deployment locations The distance between them; Represents task cluster The first in the order of execution One placement location; Define the path ratio as: ， in, Used to measure the dispersion of each AUV mission path; Definition of the first The mission payload of each AUV is: ， Define the load occupancy rate as: ， Let the path balance threshold be The load balance threshold is When satisfied ,and At that time, the task cluster adjustment is triggered; Let the current overloaded cluster be: ， in, Weighting coefficients representing path length; The weighting coefficient representing the load occupancy rate, and ; In overload clusters In the process, boundary task points are selected as candidate migration points, and for any candidate migration point... The migration cost is defined as: ， in, Indicates candidate migration points Migration to receiver cluster The resulting path increment; This represents the increase in the task payload of the AUV corresponding to the receiving cluster after migration; Indicates candidate migration points Coverage loss or cluster structure disturbance caused by removal from the original cluster; , , These are the weighting coefficients; satisfy ,and receiver cluster Comparisons are allowed, among which Indicates the load consumption of candidate migration points; Indicates receiving cluster The total length of the navigation route; Indicates the maximum allowed total voyage path length; Choose the migration plan that minimizes the migration cost: ， Candidate migration points Remove from the original cluster and reassign to the target cluster. : ， in, Indicates candidate migration points The new overloaded cluster after being removed from the original cluster; Indicates an overloaded cluster; ， in, This indicates adding candidate migration points to the original target cluster. The new target cluster; After the update, the relevant cluster centers were recalculated: ， And recalculate the path length and task load for each cluster; When satisfied and At this point, the dynamic balancing adjustment ends, resulting in the final set of task clusters: ， The final output includes the task cluster for each AUV and its corresponding deployment location set: ， in, Indicates the first The AUV ultimately executes a cluster of tasks.

Images