A multi-uuv cooperative target search method based on topological partition and soft task allocation

Abstract

This invention proposes a collaborative target search method for multiple unmanned underwater vehicles (UUVs) based on topological partitioning and soft task allocation. A search area map including islands, reefs, and ocean currents, along with a target probability map, is established. Topological regions are divided based on offshore distance fields and marked control watersheds. The probability quality, probability dispersion, and search value of each region are calculated to generate candidate search locations. The Fast Marching Method (FMM) is used to calculate the arrival time cost. Two-level soft task allocation is performed based on Unbalanced Optimal Transport (UOT), and the maximum number of search locations allocated is combined to complete the joint task allocation for multiple UUVs. Motion control actions are optimized through rolling replanning to achieve dynamic target search. This invention enables the allocation of search resources and the execution of search tasks in island, reef, and ocean current environments. In a 25-hour simulation, the weighted cumulative detection probability (WCDP) of this method is improved by 28.45% compared to the Voronoi method, thus enhancing the efficiency of multi-UUV collaborative target search.

Description

Technical Field

[0001] This invention relates to marine engineering equipment for multiple unmanned underwater vehicles (UUVs), particularly to the field of multi-UUV cooperative target search and task allocation in island, reef, and ocean current environments, and especially to a multi-UUV cooperative target search method based on topological partitioning and soft task allocation. Background Technology

[0002] In modern ocean exploration, underwater search and rescue, and marine environmental monitoring, unmanned underwater vehicles (UUVs), as underwater devices with autonomous maneuverability, sonar detection, and mission execution capabilities, have been increasingly applied to target search missions in unknown sea areas. For missions with large search areas, uncertain target locations, and the influence of the marine environment, a single UUV is limited by its detection range, speed, and mission time, making it difficult to complete a large-scale search within a limited time. Therefore, it is necessary to utilize multiple UUVs to collaboratively execute search missions, improving overall search efficiency through the rational allocation of search areas and planned navigation paths. Existing multi-UUV research also typically considers complex underwater environments, obstacles, and ocean currents as important influencing factors in collaborative path planning.

[0003] When conducting dynamic target searches in complex sea areas containing islands, narrow channels, and ambient currents, the target location exhibits significant uncertainty, and the target probability continuously changes with target movement, ocean current influences, and UUV observations. Furthermore, connectivity conditions vary between different search areas; islands and reefs can cause detours, and ambient currents can alter the actual speed of UUVs in different directions. Therefore, multi-UUV collaborative searches not only require determining search priorities based on target probability distributions but also necessitate careful consideration of the spatial structure between search areas, the actual cost of UUVs reaching the search area, and mission coordination among multiple UUVs.

[0004] Existing multi-UUV cooperative target search methods, such as probabilistic map-guided search, region decomposition, swarm intelligence optimization, and task allocation, still have the following shortcomings when dealing with this type of problem:

[0005] (1) Insufficient matching between search area division and target probability distribution: Some methods directly generate search tasks based on the location of the target with a high probability, which is easily affected by changes in local probability peaks. When the target probability map is multi-peaked or dispersed, the search tasks may change frequently. Some area division methods mainly divide the search sea area based on geometric location, without fully combining the target probability quality and probability distribution characteristics of each area, making it difficult to reasonably reflect the actual search needs of different areas.

[0006] (2) Deviation between task allocation costs and actual navigation conditions: Some existing task allocation methods mainly determine the task relationship based on the Euclidean distance or geometric path length between the UUV and the search task location. However, in complex island and ocean current environments, island obstruction will cause navigation path detours, and downstream and upstream currents will change the actual arrival time of the UUV. Using only geometric distance is difficult to accurately reflect the actual navigation costs of UUVs performing different search tasks.

[0007] (3) Uneven allocation of multi-UUV search resources: In the process of multi-UUV collaborative search, when multiple UUVs tend to the same high-value area or search task location at the same time, it is easy to cause excessive concentration of search resources, while other areas with search value may not be searched in time; for search tasks with high access costs or temporarily unreachable, traditional task allocation methods lack the corresponding adjustment capabilities, making it difficult to reasonably coordinate search needs, reachability relationships and the number of UUVs.

[0008] Therefore, there is an urgent need for a collaborative target search method that can comprehensively consider the spatial structure of the sea area, the dynamic probability distribution of targets, the actual arrival cost under ocean current conditions, and the allocation of search resources for multiple UUVs, in order to solve the problems of unreasonable search area organization, inconsistent task allocation costs with actual navigation conditions, and uneven allocation of search resources in existing technologies.

[0009] Meanwhile, compared to the present invention, the following methods differ as follows:

[0010] The paper "Multipopulation Grey Wolf Optimization for Cooperative Multi-AUV Path Planning in Complex Underwater Environments" proposes a multipopulation grey wolf optimization method for cooperative multi-AUV path planning in complex underwater environments. This method improves multi-AUV path planning capabilities by refining the swarm intelligence optimization process and considers environmental factors such as obstacles and ocean currents. However, this method primarily addresses multi-AUV navigation path optimization problems, with less consideration given to regional aggregation of dynamic target probability maps and hierarchical task allocation for search tasks.

[0011] The paper "A novel robust algorithm for path planning of multiple autonomous underwater vehicles in the environment with ocean currents" addresses the path planning problem of multiple AUVs in ocean current environments. It establishes a robust path planning model for multiple AUVs considering ocean current uncertainties to improve path planning results under varying ocean current conditions. While this method focuses on the impact of ocean currents on the navigation paths of multiple AUVs, it does not further incorporate dynamic target probability distributions or hierarchical allocation of search resources for different search areas and mission requirements.

[0012] The paper "An environment information-driven online bi-level path planning algorithm for underwater search and rescue AUV" proposes an environment information-driven online bi-level path planning method. This method combines global and local path planning to adjust the AUV search path in real time based on changes in environmental information. While this method primarily addresses the online path planning problem in underwater search and rescue operations involving a single AUV, it does not specifically address the allocation of search areas, task reachability, and coordination of search resources among multiple AUVs.

[0013] The patent "A Multi-UAV Cooperative Search Method Based on Dynamic Region Division and Multi-Constraint Virtual Force Field" proposes a multi-UAV cooperative search method based on dynamic region division. This method dynamically divides the task area using a weighted Voronoi diagram and updates the region weights based on UAV position, search progress, and environmental complexity. Within each responsibility region, a comprehensive utility function is used to select task points in areas with insufficient coverage. Then, target attraction, exploration drive, obstacle avoidance force, and region boundary constraints are used to generate the UAV's motion direction. This method is primarily aimed at multi-UAV area coverage search, where the value of task points is mainly determined by factors such as coverage progress, distance, and environmental complexity. However, when the dynamic target probability is multi-peaked or dispersed, and when there are significant differences in air traffic conditions across different areas, it still struggles to simultaneously coordinate area search needs, actual air traffic costs, and the allocation of multi-UUV search resources.

[0014] Existing technologies have achieved certain research results in target probability map-guided search, path planning in complex environments, and multi-UUV collaborative planning. However, they still have shortcomings in considering the spatial structure of the search area, dynamic target probability distribution, actual arrival time cost, and joint consideration of multi-UUV search resource allocation in complex island and ocean current environments. To address the problems of unreasonable search area organization, inconsistencies between task allocation costs and actual navigation conditions, and insufficient coordination capabilities of multi-UUV search resources in existing technologies, this invention proposes a multi-UUV collaborative target search method based on topological partitioning and soft task allocation. Summary of the Invention

[0015] This invention aims to address the problems of insufficient matching between search area division and target probability distribution, the inaccuracy of traditional geometric distance in reflecting actual arrival costs, and unreasonable allocation of multi-UUV search tasks during multi-UUV collaborative target search in complex island, reef, and ocean current environments. To this end, a multi-UUV collaborative target search method based on topology partitioning and soft task allocation is proposed. Through topology region partitioning, region probability aggregation, arrival time cost calculation, and two-level soft task allocation, the collaborative allocation and rolling execution of multi-UUV search tasks are achieved.

[0016] Specifically, the following steps are included:

[0017] Step 1:

[0018] Establish a two-dimensional grid search area that includes island and reef obstacles and environmental ocean currents, and denote the obstacle grid set as... The free water grid set is denoted as Initialize the UUV's position, heading, still water speed, motion constraints, and sonar detection parameters, and establish a target probability map in free water. .

[0019] Based on target movement trends, ocean current effects, target movement uncertainties, and multi-UUV sonar observations, the target probability map is predicted and updated using Bayesian methods.

[0020] Step 2:

[0021] Calculate the offshore distance field of the free water area based on the barrier grid. Let... For free water grids, For the obstacle grid, its distance from the shore for:

[0022]

[0023] In the formula, A set of obstacle grids; A set of free water grids; It is a two-dimensional Euclidean norm.

[0024] Regional markers are generated based on local maxima in the offshore distance field. A marker-controlled watershed is then applied to the negative offshore distance field, ensuring that watershed boundaries form in narrow channels or locations with small offshore distances, thus obtaining a set of topological regions. .

[0025] Step 3:

[0026] The target probability map is aggregated according to topological regions. Let the first... The topological regions are Its regional target probability quality for:

[0027]

[0028] when At that time, the probability of targets within the region is normalized, and the probability dispersion is calculated based on the normalized probability. :

[0029]

[0030] In the formula, For the first The number of free water grids contained in each topological region; For the first in this region The normalized probability of each grid cell.

[0031] Based on the probability quality of regional targets and probability dispersion Calculate the search value of the region :

[0032]

[0033] In the formula, This represents the probability dispersion weight. Candidate search regions within the current global replanning cycle are selected based on the probability quality of the regional target. The candidate search regions are the topological regions to be searched selected from the set of topological regions within the current global replanning cycle.

[0034] Step 4:

[0035] The number of candidate search locations is determined based on the area and probability dispersion of the candidate search region. Candidate search locations are selected from the local maxima of the target probability within the candidate search region, and the free water area within the candidate search region is divided into corresponding candidate search sub-regions.

[0036] Let the first Within the candidate search area, the first The target probability quality of each candidate search sub-region is The candidate search position corresponding to the candidate search sub-region with the highest target probability quality is selected as the representative search position of the region. :

[0037]

[0038] In the formula, For the first Within the candidate search area, the first One candidate search location; This represents the number of candidate search locations within the region. This is the candidate search position number corresponding to the candidate search sub-region with the highest target probability quality.

[0039] Step 5:

[0040] The arrival time field is calculated using the current location of each UUV as the source point and the rapid travel method that takes into account the influence of environmental ocean currents.

[0041] Let the first The arrival time field of the UUV is , No. The region representing the search location of each candidate search region is: Candidate search positions are The time cost of arrival in the region and the time cost of arrival in the candidate search location are respectively:

[0042]

[0043] In the formula, For the first UUV to the Each region represents the time cost of reaching the search location; For the first The arrival time cost of a UUV to a candidate search location.

[0044] The reachability relationship between UUVs and each search location is determined based on the arrival time cost.

[0045] Step 6:

[0046] Regional soft tasks are allocated based on regional search value and regional arrival time cost. Then, candidate search location soft tasks are allocated based on the regional soft task allocation results, the probability quality of candidate search sub-region targets, and the arrival time cost of candidate search locations.

[0047] Both levels of soft task allocation employ an unbalanced optimal transmission model. Let... Assign levels to soft tasks. This is the arrival time cost matrix for the corresponding level. The optimal soft task allocation result is obtained by assigning a weight matrix to the soft tasks:

[0048]

[0049] In the formula, For the first Layer soft task allocation results; The entropy regularity coefficient; This is the entropy regularization term; This is a relaxation term for the edge constraints on the UUV side and the mission side.

[0050] The soft task allocation results of candidate search positions within each candidate search region are merged to obtain the soft task allocation weight matrix between UUV and all candidate search positions. .

[0051] Step 7:

[0052] The total allocation weight of each UUV is calculated based on the soft task allocation results, and the UUVs participating in task allocation are determined based on the task participation threshold.

[0053] Let the first UUV and the first The soft task allocation weights among the candidate search positions are: The total weight it obtained is Then, the normalized task preference of UUV for candidate search locations for:

[0054]

[0055] The maximum number of search locations to be allocated is determined based on the allocation weights obtained for the candidate search locations and the number of UUVs that can reach those search locations. .

[0056] Step 8:

[0057] A multi-UUV joint task allocation model is established based on UUV task preferences, the maximum number of search locations to be allocated, and the reachability of candidate search locations.

[0058] set up Assign variables to the joint task. Assign matrices to joint tasks. To determine the feasible allocation set that satisfies the constraints of single UUV single task, maximum number of search location allocations, and reachability, then:

[0059]

[0060] In the formula, The optimal joint task allocation result; The set of UUVs participating in task allocation; when the UUVs were allocated to the first When there are candidate search locations, ,otherwise .

[0061] The current task search location of each UUV is determined based on the joint task allocation results.

[0062] Step 9:

[0063] Generate UUV motion control actions based on the current task search location. Let the... The increment of the UUV's steering angle within the current time step is The forward propulsion distance is Then its motion control action for:

[0064]

[0065] Construct a remaining arrival time field with the current mission search location as the endpoint, optimize motion control actions based on the remaining arrival time field, and verify the safety distance between the mission sea area boundary, island and reef obstacles, and UUVs for the candidate flight segments. Execute motion control actions that meet the constraints and update the UUV status.

[0066] Step 10:

[0067] After completing the initialization in step 1 and the topology region division in step 2, the target probability map update in step 1 and steps 3 to 9 are executed cyclically according to the rolling replanning method until the preset search time or search termination condition is reached, and the multi-UUV collaborative search path is output.

[0068] The present invention has the following beneficial effects:

[0069] (1) The method of this invention can reasonably determine the search area based on the spatial structure and target probability distribution of the search sea area, thereby improving the targeting of multiple UUVs to high-value areas. This invention performs topological division of free waters containing islands and reefs through offshore distance fields and marker control watersheds. On this basis, the target probability map is aggregated according to topological regions, and the regional search value is calculated through regional target probability quality and probability dispersion. This makes the search task no longer directly dependent on a single high-probability grid, but rather uses topological regions with spatial connectivity as the basis for organizing the search task. When the target probability distribution changes locally, the search area division can be kept relatively stable through regional probability aggregation. At the same time, the search priority can be adjusted according to the probability quality and dispersion within the region, avoiding frequent shifts in the search task due to changes in local probability peaks.

[0070] (2) The method of this invention enables the multi-UUV task allocation results to better conform to the actual navigation conditions in island and ocean current environments, and reduces the problem of uneven distribution of search resources. This invention uses a rapid travel method that considers the influence of environmental ocean currents to calculate the arrival time cost of UUVs to the regional representative search position and candidate search positions, so that the task allocation process simultaneously considers the impact of island and reef detours, search position accessibility, and downstream and upstream conditions on actual navigation time. On this basis, the matching relationship between UUVs and search tasks is determined step by step through regional soft task allocation and candidate search position soft task allocation, and the maximum number of search positions allocated is determined according to the soft task allocation weight, the accessibility of candidate search positions, and the number of UUVs participating in the search, further completing the multi-UUV joint task allocation, thereby avoiding multiple UUVs from being concentrated on a few high-probability candidate search positions at the same time, and taking into account regional search needs, navigation costs, and search resource allocation.

[0071] (3) The method of this invention can continuously adjust the search task and movement direction according to the changes in target probability distribution and UUV status, improving the adaptability of multi-UUV collaborative search in dynamic environments. The current search position of each UUV is determined by the joint task allocation results, and motion control actions are generated and optimized based on the corresponding arrival time field. Candidate routes are checked for island and reef obstacles, task sea area boundaries, and safe distances between UUVs. During the search process, a rolling replanning method is used to periodically update the target probability map, regional search value, candidate search positions, arrival time costs, and task allocation results, ensuring that the search observations of the previous cycle continue to influence subsequent search decisions. This forms a closed-loop search process consisting of target probability updates, search area organization, task allocation, and UUV motion execution, reducing ineffective navigation and unreasonable task allocation, and improving the overall efficiency and path rationality of multi-UUV collaborative target search in complex sea areas. Simulation results show that the WCDP of the method of this invention is 0.9862 at 25 hours, while the WCDP of the Voronoi method is 0.7678, representing an improvement of approximately 28.45%. Attached Figure Description

[0072] Figure 1 This is a flowchart of a multi-UUV collaborative target search method based on topology partitioning and soft task allocation as described in this invention;

[0073] Figure 2 The simulation diagram shows the search trajectories of 4 UUVs searching for 20 dynamic targets at different times using the method of the present invention.

[0074] Figure 3 Simulation diagrams of the search trajectories of 20 dynamic targets by 4 UUVs using the Voronoi method at different times;

[0075] Figure 4 This is a graph showing the weighted cumulative detection probability over time for the method of this invention and the Voronoi method. Detailed Implementation

[0076] Figure 1 This is a flowchart of a multi-UUV collaborative target search method based on topology partitioning and soft task allocation according to the present invention, including the following steps:

[0077] Step 1:

[0078] The search area is set to have both horizontal and vertical lengths. Divide it into A two-dimensional raster map. The raster cells occupied by islands and reefs are set as a set of obstacle raster cells. The set of free-water grids in which UUVs can navigate is denoted as Set the number of UUVs to [number]. Ships, number of dynamic targets The total simulation time is [number] times. The time step is The total number of time steps is:

[0079]

[0080] In the formula, This represents the total number of simulation time steps. The global replanning period is set to [value]. That is, the topological region attributes, arrival time cost, and task allocation results are recalculated every 4 time steps.

[0081] Set the maximum still water velocity of the UUV to The maximum effective radius of sonar is The maximum steering angle within a single time step is The maximum forward propulsion distance within a single time step is:

[0082]

[0083] The range of the dynamic target's own motion speed is: ,unit Its movement is driven by both its own random maneuvering and the surrounding ocean currents.

[0084] Step 2:

[0085] Initialize the motion state and target probability map for each UUV. UUV at time The state is denoted as:

[0086]

[0087] In the formula, and They represent the first The location coordinates of the UUV This indicates its current course.

[0088] Four UUVs were positioned at different boundary locations within the search area, with their initial positions within free water. If an initial position fell into an obstacle zone, it was adjusted to the nearest safe free grid.

[0089] Establish an initial target probability map on all free water grids. Each free grid cell is initialized with uniform probability:

[0090]

[0091] In the formula, Represents a set of free water grid cells. Indicates the number of free water grid cells.

[0092] Step 3:

[0093] At the beginning of each time step, the instantaneous sonar detection probability of each UUV to the free grid is calculated based on its position before movement. UUVs against grid The instantaneous detection probability is:

[0094]

[0095] In the formula, For the first The location of the UUV ,Right now The range attenuation parameter is the probability of sonar detection. This is the line-of-sight indicator. When the line connecting the UUV and the grid crosses islands or reefs... Set to 0; otherwise set to 1.

[0096] Multiple UUV grid The overall instantaneous detection probability is:

[0097]

[0098] The target probability map from the previous time step is sequentially subjected to target motion probability transfer, probability migration under ocean current influence, and prediction bias probability diffusion to obtain the predicted probability map before the observation update.

[0099]

[0100] In the formula, This represents the probability transition caused by the target's own motion. This represents the probability diffusion caused by environmental ocean currents. This represents the probability diffusion caused by the uncertainty of the target's motion.

[0101] Set the probability on the obstacle grid to 0 and perform a Bayesian update based on sonar observations that did not form effective detections:

[0102]

[0103] In the formula, After completing target motion prediction, ocean current analysis, and probability diffusion, the target is located in the grid. The predicted probability; The updated target probability based on sonar observations; The free water grid index used for normalization summation. The updated target probability map satisfies the normalization requirements within the free water area:

[0104]

[0105] Cumulative Detection Probability (CDP) and weighted cumulative detection probability are defined as metrics for result analysis. Let the... A dynamic objective at a historical moment The position is Then the dynamic target ends at time Cumulative detection probability for:

[0106]

[0107] In the formula, For UUV at time The overall instantaneous detection probability of this dynamic target.

[0108] Then time The weighted cumulative detection probability is:

[0109]

[0110] In the formula, For a moment The weighted cumulative detection probability; The number of dynamic targets.

[0111] During the search process, the cumulative detection probability and weighted cumulative detection probability are updated step by step. At each time step, based on the current position of each dynamic target and the combined instantaneous detection probability of multiple UUVs, the cumulative detection probability of each dynamic target is updated, and the weighted cumulative detection probability at the current time is calculated, thus obtaining the change process of the weighted cumulative detection probability with search time.

[0112] Step 4:

[0113] For any free water grid Calculate the minimum Euclidean distance from it to the set of obstacle grids to obtain the offshore distance:

[0114]

[0115] In the formula, Indicates the center position of the obstacle grid. This indicates the distance from the grid to the nearest island or unnavigable boundary. This represents the two-dimensional Euclidean norm.

[0116] Local maxima are selected in the offshore distance field as regional markers, and the negative offshore distance field is further analyzed. Implement marked control watershed demarcation. Since the center of open waters is farther from the shore than narrow channels and island edges, watershed boundaries preferentially form at channel contractions. Perform connectivity checks on the demarcated regions. If a region contains multiple unconnected free water portions, it is split according to four-neighbor connectivity. The final set of topological regions is denoted as:

[0117]

[0118] In the formula, This represents the total number of topological regions. For the first A topological region.

[0119] Step 5:

[0120] Calculate the regional target probability quality and probability dispersion for each topological region based on the current target probability map. For the first... Topological regions Its regional target probabilistic quality is:

[0121]

[0122] In the formula, Indicates time No. The probability quality of regional targets in a topological region Indicates time The target is located in the free grid. The probability of.

[0123] when When calculating the normalized probability within the region:

[0124]

[0125] In the formula, Indicates the first Free grid within a topological region The normalized probability.

[0126] set up For the first The number of free water grid cells contained in each topological region. For the first in this region The normalized probability of each grid cell is then given by the probability dispersion:

[0127]

[0128] In the formula, Indicates the first The probability dispersion of a topological region. When or season ;when When the entropy term is zero, the corresponding term is 0.

[0129] Calculate the region search value based on the region's target probability quality and probability dispersion:

[0130]

[0131] In the formula, Indicates the first The regional search value of a topological region This represents the probability dispersion weight.

[0132] Set the regional target probability quality threshold to 0.01. If the topological region is selected as a candidate search region in the current global replanning cycle, then the region will be selected as a candidate search region in the current cycle; otherwise, no search task will be generated for the region in the current cycle, and the region will be re-evaluated in the next global replanning cycle.

[0133] Step 6:

[0134] No. The area of ​​each candidate search region is The number of candidate search locations is determined by combining the region area and probability dispersion.

[0135]

[0136] In the formula, For the first The number of candidate search locations requested for each candidate search region; and These represent the lower and upper limits of the number of candidate search positions, respectively. The unit candidate area is used to limit the scale of the region represented by a single candidate search location.

[0137] The minimum spacing between candidate areas and candidate search positions are set as follows:

[0138]

[0139] In the formula, The unit candidate area adjustment coefficient; This is the minimum spacing between candidate search positions, used to avoid multiple candidate search positions being overly concentrated near the same probability peak.

[0140] From the local maxima of target probability within the candidate search region, grid cells satisfying the minimum spacing requirement are selected sequentially as candidate search locations. If the actual number of obtained candidate search locations is less than... If a candidate is found, the existing candidate is retained; if no candidate search location is found, the free grid with the highest target probability within the area is selected as the candidate search location.

[0141] Let the first The set of candidate search locations generated within each candidate search region is The actual number of candidate search locations is denoted as Each free grid cell within the candidate search area is assigned to the nearest candidate search location:

[0142]

[0143] In the formula, Represents a grid The candidate search sub-region to which it belongs. Indicates the first Within the topological region, the first The first candidate search location. The target probability quality of each candidate search sub-region is:

[0144]

[0145] In the formula, For the first Target probability quality of each candidate search sub-region This is an indicator function.

[0146] The candidate search position with the highest target probability quality in the candidate search sub-region is selected as the first... Each topological region represents a search location. :

[0147]

[0148] In the formula, This is the corresponding candidate search position number.

[0149] Step 7:

[0150] Based on the current location of each UUV, the FMM arrival time field considering the influence of ocean currents is solved according to the obstacle grid, the still water velocity of the UUV, and the ambient ocean currents. Let... For global replanning time Ocean current velocity vector, For the direction of propagation of a unit, the first UUV along direction The speed under the ocean current is:

[0151]

[0152] In the formula, For the first UUV in grid Along the direction of unit propagation Speed ​​under ocean currents; For the first The still water speed of a UUV; For global replanning time The ocean current velocity vector; The direction of propagation is defined by the unit. Arrival time field of UUV satisfy:

[0153]

[0154] In the formula, Indicates the first UUV from current location Reaching the free grid The shortest time required The spatial gradient of the arrival time field. The barrier grid prohibits arrival time propagation.

[0155] Arrival time fields are sampled at the representative search location and candidate search location in the region to obtain the region arrival time cost and candidate search location arrival time cost, respectively:

[0156]

[0157] In the formula, For the first UUV to the Each region represents the time cost of reaching the search location. For the first UUVs to candidate search locations The arrival time cost. If the corresponding location is unreachable, the corresponding relationship is marked as unreachable.

[0158] Step 8:

[0159] Two levels of soft task allocation are performed based on regional search value and arrival time cost.

[0160] The first level is the allocation of soft tasks in the region. In this embodiment, each UUV adopts a homogeneous configuration, so the UUV side adopts an equal weight setting; the regional task weight distribution is constructed according to the regional search value of each candidate search region, and the regional arrival time cost is used as the matching cost to solve the soft task allocation weight between UUV and candidate search regions.

[0161] The second level involves soft task allocation for candidate search locations within the region. Based on the soft task allocation results, the set of UUVs participating in the search of each candidate search region is determined. A UUV-side weight distribution is constructed based on the regional allocation results of the participating UUVs. A search location-side task weight distribution is constructed based on the target probability quality of each candidate search sub-region. The soft task allocation weights between UUVs and candidate search locations are solved using the arrival time cost of candidate search locations as the matching cost.

[0162] The two levels of soft task allocation described above employ the same entropy-regularized unbalanced optimal transmission model. For any allocation level... Its objective function is:

[0163]

[0164] In the formula, For the first Layer soft task allocation results For the first The arrival time cost matrix of the layer The entropy regularity coefficient is... For the first The entropy regularization term of the layer's soft task assignment weight matrix. For the first Relaxation terms for the UUV side and mission side edge constraints of the layer. Let... The entropy regularization term is:

[0165]

[0166] The marginal relaxation term is:

[0167]

[0168] In the formula, and The first UUV side weight distribution and task side weight distribution of the layer; and These are the edge relaxation coefficients for the UUV side and the mission side, respectively. The divergence is Kullback–Leibler.

[0169] The generalized Sinkhorn iteration is used to solve the two-level soft task allocation problem, and the soft task allocation results of candidate search positions in each candidate search region are merged to obtain the soft task allocation weight matrix between UUV and all candidate search positions. For the relationship between unreachable UUVs and candidate search locations, the corresponding assigned weight is set to 0.

[0170] Step 9:

[0171] The maximum number of search locations to be allocated is determined based on the soft task allocation weights, and the multi-UUV joint task allocation is completed. The calculation of the... Total allocation weights obtained by each UUV:

[0172]

[0173] In the formula, For the first The total weight a UUV receives in the soft task assignment of candidate search locations.

[0174] Set task participation threshold .when When this happens, the UUV is identified as a UUV participating in task allocation. The set of UUVs participating in task allocation is denoted as . Its quantity is recorded as The allocation weights of the UUVs participating in task allocation are normalized to obtain their task preferences for each candidate search location:

[0175]

[0176] In the formula, For the first UUV selection Normalized task preference for candidate search locations. The weighting ratios assigned to each candidate search position are:

[0177]

[0178] In the formula, Indicates the first The weighting of each candidate search position within the total number of candidate search positions. Let... In order to reach the first Given the number of UUVs with positive assignment weights at each candidate search position, the initial maximum number of assignments for each candidate search position is:

[0179]

[0180] In the formula, For the first The initial maximum number of allocations for each candidate search location represents the number of UUVs allowed to be allocated to that candidate search location within the current global replanning cycle; This is to ensure that the maximum number of search locations allocated does not exceed the actual number of UUVs that can reach that search location.

[0181] When the sum of the maximum allocation quantities of all candidate search positions is less than the number of UUVs participating in the task allocation, the maximum allocation quantity of each candidate search position is gradually increased according to the remaining allocation weight ratio of each candidate search position until the joint task allocation requirements are met.

[0182] set up Assign variables to the joint task. When the first... The UUV was assigned to the first When there are candidate search locations, ;otherwise, The multi-UUV joint task allocation model is as follows:

[0183]

[0184] In the formula, Assign matrices to joint tasks. To achieve the optimal joint task allocation result, This constitutes a feasible allocation set. The feasible allocation set satisfies the following conditions: each UUV participating in the task allocation receives only one candidate search location; allocation to the [number missing] [unit missing]... The number of UUVs at each candidate search location does not exceed its maximum allocation. ;when or season ;and .

[0185] This embodiment employs deterministic subset dynamic programming to solve the aforementioned joint task allocation model, and determines the candidate search positions allocated to each UUV as the current task search positions. UUVs that have not obtained a current task search position remain in a safe waiting state and re-participate in task allocation in the next global replanning cycle.

[0186] Step 10:

[0187] Generate UUV motion control actions based on the current task search location. The motion control actions of a UUV consist of steering angle increment and forward thrust distance:

[0188]

[0189] In the formula, For the first Motion control actions of a UUV This represents the increment of the steering angle within the current time step. This represents the forward propulsion distance. The candidate action endpoint is obtained based on the motion control action.

[0190]

[0191] In the formula, The endpoint of the candidate action; This is the current location of the UUV; This represents the forward propulsion distance of the UUV within the current time step. This represents the ocean current velocity vector at the current location; Let be the time step. Let the be the... The current mission search location of the UUV is And construct the remaining arrival time field with the current task search position as the endpoint. ,in Indicates the location of free water. Go to the current task search location The remaining arrival time. Based on the remaining arrival time field, candidate motion control actions within the allowable range are evaluated:

[0192]

[0193] In the formula, This is the evaluation value for motion control actions; End point of candidate action To the The current search location of the UUV The remaining arrival time; To reduce the penalty weight for turning, and to avoid unnecessary large-angle turns by UUVs; Obstacles and boundary safety penalties for candidate flight segments; Weighting for security penalties.

[0194] The particle swarm optimization algorithm is used to find the candidate motion control action with the minimum evaluation function within the feasible region of motion control action.

[0195] Step 11:

[0196] Perform segment safety checks on motion control actions and update UUV positions. Each motion control action corresponds to a candidate segment. Set equidistant points along the candidate segments. The sampling point, the first The number of sampling points is:

[0197]

[0198] In the formula, This refers to the number of sampling points for a flight segment, used to check whether a candidate flight segment crosses islands or reefs, exceeds the mission area, or does not meet the safe distance between UUVs. The first on the candidate flight segment One sampling point. The distance between adjacent sampling points is determined based on the candidate flight segment length and grid spacing, so that the distance between them can meet the obstacle crossing inspection requirements.

[0199] If all sampling points meet the constraints, the selected motion control action is executed and the UUV position is updated; if the candidate segment goes beyond the mission area, crosses islands and reefs, or does not meet the safe distance constraints, the UUV is retreated to the last safe position on that segment.

[0200] Step 12:

[0201] The above search process is repeated using a rolling replanning method.

[0202] At the initial time step and at each global replanning time step, based on the latest target probability map, UUV location, and environmental ocean current state, the regional target probability quality, probability dispersion, regional search value, candidate search location, regional representative search location, arrival time cost, two-level soft task allocation results, and multi-UUV joint task allocation results are recalculated.

[0203] Within the same global replanning cycle, each UUV continuously generates and executes motion control actions according to its current task search position. Topology regions that do not generate search tasks in the current cycle and UUVs that have not obtained their current task search positions will re-participate in task allocation in the next global replanning cycle.

[0204] Repeat the above steps until the preset search time or search termination condition is reached, and output the collaborative search path of each UUV.

[0205] To verify the search performance of the method of this invention, under the same search area, island and reef environment, ocean current field, initial state of UUV, dynamic target trajectory, sonar parameters, and search duration, the method of this invention is compared with the Voronoi method. This embodiment compares the Voronoi method of the patent "A Multi-UAV Cooperative Search Method with Dynamic Region Division and Multi-Constraint Virtual Force Field".

[0206] The method of this invention was simulated using MATLAB software. Figure 2 This is a simulation diagram of the search trajectories of 4 UUVs searching for 20 dynamic targets at different times in the method of this invention. Figure 3 Simulation diagrams of the search trajectories of 20 dynamic targets by 4 UUVs in the Voronoi method at different times. Figure 4 This is a graph showing the weighted cumulative detection probability over time for the method of this invention and the Voronoi method. From... Figure 2It can be intuitively seen that in the method of this invention, the search paths of each UUV can avoid island and reef obstacles and expand to different search areas. After searching different areas in the early stage, it can also search the searched areas again, avoiding the situation where dynamic targets are missed and enter the searched areas. During the search process, there is no phenomenon of multiple UUVs concentrating in the same local area for a long time. The search tasks among the UUVs are reasonably allocated, and the search paths can form good coordination. Figure 3 While the Voronoi method in [the original text] can perform basic cooperative search, its cumulative detection performance for dynamic targets within the same search time is lower than that of the method in this invention. Figure 4 It can be seen that at 5 hours, the weighted cumulative detection probability of the method of the present invention reaches 0.5736, while that of the Voronoi method is 0.4257; at 15 hours, the weighted cumulative detection probability of the method of the present invention reaches 0.8409, while that of the comparative method is 0.6680; at 25 hours, the weighted cumulative detection probability of the method of the present invention reaches 0.9862, which is close to 1, while that of the comparative method is only 0.7678. The weighted cumulative detection probability of the method of the present invention at 5 hours, 15 hours, and 25 hours is improved by approximately 34.75%, 25.88%, and 28.45% respectively compared to the comparative method, with an improvement of 0.2184 at 25 hours, indicating that the method of the present invention can achieve a higher cumulative detection effect within the same search time. Specific data are shown in Table 1.

[0207] Table 1. Comparison of WCDP results between the two methods

[0208] Search time / h The method of the present invention is WCDP Voronoi method WCDP Difference relative improvement 5 0.5736 0.4257 0.1479 34.75% 15 0.8409 0.6680 0.1729 25.88% 25 0.9862 0.7678 0.2184 28.45%

[0209] Depend on Figure 4 Furthermore, it can be observed that the weighted cumulative detection probability curve of the method of this invention continues to rise in the later stages of the search. When the curve growth slows down in some search stages, the search value of the region, candidate search positions, arrival time costs, and multi-UUV task allocation results can be recalculated through rolling replanning, thereby continuing to improve the search effect on the remaining target probability region. In contrast, the curve growth of the Voronoi method is relatively slow in multiple time intervals, and its adaptability to changes in search tasks under complex island and ocean current environments is relatively insufficient. The above data indicate that the method of this invention has certain advantages in target search efficiency, cumulative detection effect, rationality of multi-UUV search resource allocation, and adaptability to complex environments, providing a new solution for multi-UUV collaborative dynamic target search under island and ocean current environments.

Claims

1. A multi-UUV collaborative target search method based on topology partitioning and soft task allocation, characterized in that, Includes the following steps: Step 1: Establish a two-dimensional grid search area that includes island and reef obstacles and environmental ocean currents, and denote unnavigable grids as the obstacle grid set. The remaining navigable grid cells are denoted as the free water grid cell set. ; Initialize the position, heading, still water speed, motion constraints and sonar detection parameters of multiple UUVs, and establish a target probability map in the free water area; Based on the target's movement trend, ocean current effects, movement uncertainty, and multi-UUV sonar observations, the target probability map is predicted and updated using Bayesian methods to obtain the target probability map at the current moment. , Indicates the current moment. Indicates the position of the free water grid; Step 2: For any free water grid Calculate its connection to the set of obstacle grids. The minimum Euclidean distance yields the offshore distance field: In the formula, Free water grid The distance from the shore; The center position of the obstacle grid; A set of obstacle grids; A set of free water grids; It is a two-dimensional Euclidean norm; Regional markers are generated based on the local maxima of the offshore distance field, for the negative offshore distance field. By implementing marker-controlled watershed delineation, watershed boundaries are formed in narrow waterways or at locations with low distances from the shore, resulting in a set of non-overlapping topological regions: In the formula, A set of topological regions; For the first One topological region; This represents the total number of topological regions. For topological region indexing, ; Step 3: Aggregate the grid probabilities in the current target probability map according to the topological region, and calculate the first... Regional target probability quality of each topological region: In the formula, For a moment No. Regional target probability quality of each topological region; For a moment The target is located in the grid. The probability of; For the first A topological region; when When calculating the normalized probability within the region: In the formula, For a moment No. Grid within a topological region The normalized probability; For the target located in the grid The current probability; For the first Regional target probability quality of each topological region; set up For the first The number of free water grids contained in each topological region is used to calculate the probability dispersion based on the normalized probability within the region: In the formula, For a moment No. The probability dispersion of each topological region; This represents the number of free water grid cells contained in this topological region; For the raster index within the region; For the first in the region The normalized probability of each grid cell is calculated; and the area search value is calculated based on the area target probability quality and probability dispersion. In the formula, For a moment No. The regional search value of a topological region; For the probability quality of regional targets; Probability dispersion; The probability dispersion weight is used to adjust the impact of the probability dispersion within a region on the region search value; topological regions whose target probability quality reaches a set threshold are identified as candidate search regions within the current global replanning cycle. Step 4: According to the The number of candidate search locations is determined by the area and probability dispersion of each candidate search region. Then, free-water grids that meet the minimum spacing requirement are selected from the local maxima of the target probability within each region as candidate search locations, resulting in the set of candidate search locations: In the formula, For the first A set of candidate search locations for each candidate search region; For the first in this region One candidate search location; This represents the number of candidate search locations within the region. ; The first Each free water grid within a candidate search area is assigned to the nearest candidate search location, forming a candidate search sub-region: In the formula, For grid In the The candidate search sub-region number corresponding to each candidate search region; For the first One candidate search location; This represents the number of candidate search locations. Indicate the candidate search position number corresponding to obtaining the minimum distance; and calculate the first... Target probability quality of each candidate search sub-region: In the formula, For the first Within the topological region, the first Target probability quality of each candidate search sub-region; This is an indicator function; it takes the value 1 if the condition inside the parentheses is true, and 0 otherwise. For grid The candidate search sub-region number; For the target located in the grid The current probability; The candidate search position with the highest target probability quality in the candidate search sub-region is selected as the first... Each candidate search region represents a search location. : In the formula, The candidate search location number that has the highest target probability quality in the candidate search sub-region; For the corresponding candidate search positions; For the first Target probability quality of each candidate search sub-region; Step 5: Based on the current location of each UUV, the obstacle grid set, the still water velocity of the UUV, and the ocean current velocity, the arrival time field is calculated using a rapid travel method that considers the influence of environmental ocean currents; UUV in grid Along the direction of unit propagation The speed of the ship under the ocean current is expressed as: In the formula, For the first UUV in grid Along the direction of unit propagation Speed ​​under ocean currents; For the first The still water speed of a UUV; For global replanning time The ocean current velocity vector; The direction of propagation for a unit; Sample the first at the region representative search location and candidate search location. Arrival time field of UUV The arrival time cost of the region and the arrival time cost of the candidate search location are obtained respectively: In the formula, Indicates a region layer. Indicates the candidate search location layer; For the first UUV to the Each region represents the time cost of reaching the search location; For the first UUV to the Within the candidate search area, the first The arrival time cost of each candidate search location; For the first Arrival time of the UUV; The region represents the search location; Candidate search locations; And based on whether the arrival time cost is finite, the reachability relationship between UUV and the corresponding regional representative search location and candidate search location is determined; Step 6: Based on the regional search value and regional arrival time cost of each candidate search region, a regional soft task allocation model is established to obtain the regional soft task allocation weight between UUVs and candidate search regions. Then, based on the regional soft task allocation weight, the set of UUVs participating in the search of each candidate search region is determined. Based on the target probability quality of the candidate search sub-region, the task weight on the search location side is constructed. Using the arrival time cost of the candidate search location as the allocation cost, a candidate search location soft task allocation model is established to obtain the soft task allocation weight between UUVs and candidate search locations. Both the regional soft task allocation and the candidate search location soft task allocation adopt an entropy-regularized unbalanced optimal transmission model for any allocation level. The soft task allocation result is as follows: In the formula, Allocate hierarchical indexes for soft tasks, where This represents the regional soft task allocation layer. This represents the soft task allocation layer for candidate search positions; For the first The non-negative soft task weighting matrix to be solved in the layer; For the first Optimal soft task allocation results at each layer; For the first Layer arrival time cost matrix; The weighted sum of arrival time cost and allocation weight; The entropy regularity coefficient; This is the entropy regularization term; The relaxation terms for the edge constraints on the UUV side and the task side are used; the soft task allocation results of the candidate search positions in each candidate search region are merged to obtain the soft task allocation weight matrix between the UUV and all candidate search positions. , This represents the candidate search location layer, and the weights assigned to unreachable relationships are set to 0. For the first UUV and the first Soft task weighting is assigned among the candidate search locations; Step 7: According to the soft task allocation weight matrix Calculate the first Total allocation weights obtained by each UUV: In the formula, For the first The total allocation weight obtained by each UUV; For the first UUV and the first Soft task weighting is assigned among the candidate search locations; Index for candidate search locations; The UUVs whose total assigned weights reach the task participation threshold are identified as participating UUVs in task allocation, and their assigned weights are normalized to obtain the [number of UUVs]. UUV against the first Task preference for candidate search locations: In the formula, For the first UUV against the first Normalized task preference for candidate search locations; Assign weights to the corresponding soft tasks; For the first The total allocation weight obtained by each UUV; The maximum number of search positions to be allocated is determined based on the weight ratio of each candidate search position and the number of UUVs that can reach the corresponding candidate search position. This ensures that the number of UUVs assigned to the same candidate search location does not exceed the maximum number of UUVs assigned to that search location. Step 8: set up Assign variables to the joint task, when the first The UUV was assigned to the first When there are candidate search locations, ,otherwise Each UUV participating in the mission allocation will receive only one candidate search location, which will be assigned to the [number]th [location]. The number of UUVs at each candidate search location does not exceed its maximum allocation. Unreachable relationships do not participate in the distribution and The feasible assignment set is formed by assigning all joint task matrices of binary variables. And solve: In the formula, Assign matrices to joint tasks; The optimal joint task allocation result; To satisfy the single UUV single task constraint, the maximum number of search locations to be allocated, the reachability constraint, and the binary value constraint, a feasible allocation set is required; The set of UUVs participating in task allocation; Assign variables to the joint task; For the first UUV against the first Normalized task preference for candidate search locations; The current task search position of each UUV is determined based on the optimal joint task allocation result. UUVs that have not obtained a current task search position shall remain in a safe waiting state during the current global replanning cycle. Step 9: For obtaining the current task search location, the first UUVs, constructed with incremental steering angle and forward propulsion distance Composition of motor control actions: In the formula, For the first Motion control actions of a UUV; This represents the increment of the UUV's steering angle within the current time step; Forward thrust distance; Construct a remaining arrival time field with the current task search position as the endpoint, and optimize the motion control action according to the remaining arrival time field and UUV motion constraints to obtain candidate action endpoints; Sampling points are set along the candidate route from the current position of the UUV to the end point of the candidate action. The boundary of the mission sea area, island and reef obstacles and safe distance between UUVs are checked for the candidate route. Motion control actions that meet the constraints are executed and the UUV status is updated. For candidate routes that do not meet the constraints, the route is backed to the last safe position. Step 10: After completing the initialization in step 1 and the topology region division in step 2, the target probability map update in step 1 and the motion control in step 9 are executed at each time step. At each global replanning time, steps 3 to 8 are executed to update the current task search position of each UUV. The above process is repeated until the preset search time or search termination condition is reached, and the multi-UUV cooperative search path is output.

Images