A task area division and clustering method for collaborative search of multiple underwater robots

Abstract

The present invention provides a task area division and clustering method for collaborative search of multiple underwater robots, and the present invention relates to the technical field of collaborative search of underwater robots. The present invention provides a preliminary grid division of the task area based on the UUV operation width; weighting the importance of each grid in the task area; selection of the largest area inside the boundary of the task area and a clustering method of grids with similar importance; refinement of edge areas; classification of aggregated areas based on geometric features and the induction of corresponding search path planning strategies. The computational amount of the grid division of the present invention is controllable. For any task area, as long as the grid interval / width is determined, the gridding process is simple and stable, and its computational amount is also roughly proportional to the refinement of the grid division. In the method used by the present invention, its computational amount is usually always small.

Description

Technical Field

[0001] The invention relates to the technical field of underwater robot collaborative search, and in particular to a task area division and clustering method for collaborative search of multiple underwater robots. Background Art

[0002] At present, the detection and search of the mission sea area are mostly carried out by using multi-acoustic sensor detection with underwater robots as carriers. However, as the mission sea area expands to the deep sea, a single UUV equipped with multiple acoustic sensors for search is difficult to adapt to the mission requirements in terms of endurance and efficiency. The collaborative search technology of multiple underwater robots has attracted widespread attention due to its wide coverage, high search efficiency, and high detection accuracy. For specific sea areas in the deep sea, the conventional cluster search method is that multiple UUVs search the mission area in a certain formation (triangular formation distribution or horizontal line distribution). However, the current search of the target sea area is often a rough area marked out by a UUV cluster for coverage or random search, which makes it difficult to balance efficiency and accuracy.

[0003] By dividing the task area based on different importance and reasonably allocating the task sea area for the collaborative search of multiple underwater robots, the efficiency and effect of the search process can be taken into account. Therefore, designing a task area division method based on regional importance to provide a reasonable division and allocation strategy for the multi-UUV collaborative search task area is an effective technical means for multiple underwater robots to carry out collaborative search tasks. Summary of the invention

[0004] The purpose of the present invention is to provide a task sea area division method with good versatility, strong scalability and simple solution, which is used for static task allocation during collaborative search of multiple underwater robots. The present invention also includes an adaptive task area clustering method, which simplifies the path planning complexity of the search area by clustering task areas of similar importance, and improves the task execution efficiency during collaborative search.

[0005] A multi-UUV task area division method based on regional importance includes: 1. a task area gridding method based on the UUV operation width; 2. an importance weighting for the task area grid; 3. a method for selecting the largest internal area and clustering areas with similar importance; 4. a method for fine-tuning edge areas; 5. (geometric) features of different types of aggregation areas and search path planning strategies.

[0006] The present invention provides a task area division and clustering method for collaborative search of multiple underwater robots, and the specific scheme is as follows:

[0007] A task area division and clustering method for collaborative search of multiple underwater robots includes the following steps:

[0008] Step 1: Preliminary grid division of the mission area based on the UUV operation width;

[0009] Step 2: Assign weights to the importance of each grid in the task area;

[0010] Step 3: Selection of the largest area within the task area boundary and clustering method of grids with similar importance;

[0011] Step 4: Refined processing of edge areas;

[0012] Step 5: Classification of aggregation areas based on geometric features and induction of corresponding search path planning strategies.

[0013] Preferably, the step 1 is specifically:

[0014] S1.1: Adaptively configure the unit length of the grid according to the operating width of the UUV performing the task, record UUV1, UUV2, UUV3, ..., UUV n The working width is W1, W2, W3, ..., W n , take the minimum working width as W0:

[0015] w0=min{w1,w2,...,w n};

[0016] S1.2: Order As a parameter for comparing the actual operating width of each UUV with the minimum operating width;

[0017] S1.3: Take The lowest common multiple of is used as the basic unit of gridding, and integer multiples w are required according to the accuracy of task allocation. m As the grid width, make a preliminary division of the task area.

[0018] Preferably, the step 2 is specifically:

[0019] S2.1: The number of features required for importance K is UUVs are heading out for exploration;

[0020] in, Indicates rounding up, P is the total number of UUVs,

[0021] The K value is weighted and expressed by the following formula:

[0022]

[0023] Among them, C is a constant based on the average number of searches per unit time in each area;

[0024] S2.2: The calculation of F value adopts fuzzy evaluation method, and the familiarity of the area is divided into three categories: familiar, general and unfamiliar, with scores of 0, 0.5 and 1 respectively. The probability of familiarity, general and unfamiliarity in the area is counted, and the proportion is multiplied by the score weight and normalized. The sum of the normalized values ​​is taken as the F value;

[0025] The A value is calculated using a fuzzy evaluation method. The evaluation of regional needs and attention is divided into five categories: special attention, attention, some need, dispensable, and no attention at all, with scores of 1, 0.75, 0.5, 0.25, and 0, respectively. The probability of each attention level is statistically analyzed, and the proportion is multiplied by the score weight and normalized. The sum of the normalized values ​​is taken as the A value.

[0026] According to the statistics of task allocation and execution process, it is the dynamic input quantity X, with the initial value X0=0;

[0027] S2.3: Simplify the process and assign only one K value to each grid area as the importance of the grid. For any grid i, according to its F i , A i , X i The value can be obtained K i , taking three different marks as examples to assign importance to each grid in the area.

[0028] Preferably, in a specific task for searching for static seabed targets, the K value is set to be dynamically adjusted with the number of targets found to achieve multiple detections of key areas, and the average number of targets in each area, T0, is taken;

[0029] K T =K+log a (T i -T0)

[0030] The value of a is assigned according to the actual situation, a>1; when T i -T0≤0, take K T =K.

[0031] Preferably, the step three is specifically:

[0032] S3.1: The method for selecting the largest internal area is to select the coordinate value (x min ,y min )(x max ,y max )(x min ,y max )(x max ,y min ) of the four grid nodes, the rectangle formed is denoted as Q, and when the number of grids surrounded by Q is the largest, it is denoted as Q max , Qmax is the largest internal area;

[0033] S3.2: Cluster and integrate the grids whose importance is within the range of K±I, where I is the inclusiveness, which is between 5%K and 25%K. The difference method is used to test the clustering of the grids with a step size of 1%K and output the results. The I value with reasonable regional division is selected as the specified parameter input value of the algorithm;

[0034] S3.3: Perform artificial secondary aggregation on the parts that can be optimized in the regional division to ensure the reasonable and efficient aggregation of the internal regions;

[0035] S3.4: For areas that are significantly larger than the average partition area (S 聚类 >3S 平均 ) is called the out-of-bounds area. Considering that it takes a long time for a single UUV to search it, it is divided into several areas of appropriate size to improve the search efficiency;

[0036] The importance of the aggregated area is the highest importance value K+I in the area, and the aggregation principle is that low importance can be aggregated to high importance, but not vice versa.

[0037] Preferably, the principle for dividing the out-of-bounds area into appropriate areas is:

[0038] Segmentation avoids increasing the complexity of search path planning within the area;

[0039] The area of ​​the divided region should be uniform and avoid large difference from the average area;

[0040] Choose a plan with fewer splits.

[0041] Preferably, to ensure the maximum internal area Q when clustering grids with similar importance max The integrity of the grid is maintained while taking into account the clustering principle of grids with strict similarity in importance; for grids exceeding Q max The boundary grid should follow a stricter rule when clustering, that is, Q max The importance of the grid outside the boundary only falls within the range of K±0.2I, the same as Q max The grids within the boundary are clustered to form erosion and separation of the edge area L.

[0042] Preferably, the step 4 is specifically:

[0043] S4.1: Q max The area enclosed by the boundary extension and the boundary of the task area S is recorded as the edge area L, Q maxThe area enclosed by the extended edge and the boundary of the task area S starts from the leftmost area, the area with the minimum coordinate value, and is sorted in a clockwise direction, and is recorded as edge areas L1, L2, L3, and L4 respectively. Each area crossing the boundary of the previous area is not counted to avoid overlap;

[0044] S4.2: The edge region L is within the task region SQ max Area outside, that is, L = SQ max , which is composed of the grids that the boundary of S passes through and its neighboring grids, and is represented by Q max separated by the borders;

[0045] S4.3: Clustering of edge region L: The actual effective task area contained in the grid where the boundary passes is different. According to the precision requirement, the grid where the boundary passes is clustered. The standard grid is divided again, and small grids whose intersection with S is empty are removed during clustering;

[0046] S4.4: For the clustering of edge regions L, except for those separated by internal aggregation regions of different importance, all are clustered by S boundary Q max The edge area L formed by the boundary i , aggregated into a stepped polygon or rectangle L i * , for the edge area L separated by the internal areas of different importance i 1 ,L i 2 ,...,L i n Aggregate into edge regions L i *1 , L i *2 ,…,L i *n , where i=1,2,3,4.

[0047] Preferably, the step five is specifically:

[0048] S5.1: For clustering areas that are rectangular or standard stepped polygons, the longest side is used as the travel direction of the UUV comb line scan, and the search is performed from each mission area. After the search is completed, it automatically switches to the longest side direction of the adjacent mission area and performs comb line scan again;

[0049] S5.2: For rectangular or standard stepped polygonal clustering regions with convex or concave grids, where For example, in areas 1 and 5, for the protruding parts, ignore their influence first, search the main area as a regular area according to the strategy of S5.1, and then search the unsearched area; for the concave parts, directly fill them as regular areas, and then search according to the strategy of S5.1;

[0050] S5.3: For cluster regions that are rectangular or standard stepped polygons with convex or concave grids, where The convex or concave grid part is regarded as another adjacent area of ​​the same importance, and the search logic is to search immediately after or before the main area of ​​the rectangular or standard stepped polygonal part; the search strategy for the main area is the same as S5.1, and after the search is completed, it automatically switches to the longest side direction of the convex or concave grid area to perform comb line scanning again, or after searching the convex or concave grid area first, it automatically switches to the longest side direction of the main area to perform comb line scanning. The search order is determined by the distance between the end point of the previous area searched by the UUV and the search starting point of the main area and the convex or concave grid area;

[0051] S5.4: The preset number of detections per unit time in each task area is M ( = represents rounding, P is the total number of UUVs), the initial detection times of each mission area is N = 0, when any UUV searches the designated mission area, the detection times in the area is N = N + 1, and the UUV will be marked; considering the same UUV's re-detection of the mission area, the actual benefit decreases, and the multiple searches of the marked UUV on the same mission area are equivalent to:

[0052] The actual number of detections in the mission area is:

[0053]

[0054] Among them, a i is the judgment function. When the number of detections of the i-th UUV is n i ≥2, a i =1; otherwise a i =0, b i is the judgment function. When the number of detections of the i-th UUV is n i ≥1, b i =1; otherwise b i =0.

[0055] Preferably, in order to prevent the situation where all UUVs near the area have completed the search but the search times have not reached the importance requirement, and the distant UUVs have to wait too long to perform the search, before the search times in the mission area reach the importance requirement, the neighboring UUVs can be continuously summoned to repeat the search until the importance requirement condition N≥M is met.

[0056] Beneficial effects:

[0057] The preliminary division of the task area of ​​the present invention adopts a gridding method, which has the following advantages:

[0058] (1) The task area is divided into relatively regular blocks and is convenient for the planning of comb lines;

[0059] (2) The search task volume of the divided blocks can be easily estimated and calculated, which is conducive to the task allocation of multi-UUV collaborative search; (The estimation can be done by comparing the grid area occupied by the blocks, calculating the comb line scanning in the direction of the long side of the block according to the UUV search width, and calculating the total stroke of the comb line)

[0060] (3) The computational complexity of grid division is controllable. For any task area, as long as the grid interval / width is determined, the gridding process is simple and stable, and the computational complexity is roughly proportional to the degree of fineness of the grid division. In the method used in the present invention, the computational complexity is usually always small.

[0061] The advantages of the grid importance weighting method in the task area of ​​the present invention are:

[0062] (1) The assignment of importance values ​​takes into account both the static input before task execution and the dynamic input during task execution. It not only refers to prior knowledge, but also continuously optimizes the importance index based on a posteriori knowledge to make it meet the task requirements.

[0063] (2) Determining the search frequency of a specific area based on its importance is of practical significance and is more in line with actual task requirements;

[0064] (3) In the design of the search strategy that meets the importance requirements, the situation where the number of neighboring UUVs is insufficient is taken into account, and a strategy of UUV repeatable search with diminishing returns is adopted. This reduces the task waiting time on the basis of considering actual needs, making the overall efficiency higher.

[0065] The grid clustering processing method in the task area of ​​the present invention has the advantages of:

[0066] (1) The clustering of grids uses mathematical processing methods with different degrees of inclusion, which can meet the accuracy requirements of actual tasks and divide regular areas;

[0067] (2) The task area is divided relatively evenly, which reduces the difficulty of task allocation and makes it easier to obtain the optimal solution in time during UUV collaborative search;

[0068] (3) Edge processing can ensure that the UUV can fully cover the mission area;

[0069] (4) Relatively reasonable search strategies are given for aggregation areas with different geometric features to reduce the difficulty of path planning and improve search efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 Gridding for the mission area;

[0071] Figure 2 Empower importance;

[0072] Figure 3 The largest rectangle Q in the selected area max ;

[0073] Figure 4 Clustering of grids with similar importance;

[0074] Figure 5 Cutting for large clustering areas;

[0075] Figure 6 is the edge area L;

[0076] Figure 7 For edge refinement and clustering;

[0077] Figure 8 Different types of aggregation areas. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] Combination Figures 1 to 8 As shown, the present invention provides a task area division and clustering method for collaborative search of multiple underwater robots, comprising the following steps:

[0080] 1. A task area division and clustering method for collaborative search of multiple underwater robots, characterized in that it includes the following steps:

[0081] Step 1: Preliminary grid division of the mission area based on the UUV operation width;

[0082] 2. The step 1 is specifically as follows:

[0083] S1.1: Adaptively configure the unit length of the grid according to the operating width of the UUV performing the task, record UUV1, UUV2, UUV3, ..., UUV n The working width is W1, W2, W3, ..., Wn , take the minimum working width as W0:

[0084] w0=min{w1,w2,...,w n};

[0085] S1.2: Order As a parameter for comparing the actual operating width of each UUV with the minimum operating width;

[0086] S1.3: Take The lowest common multiple of is used as the basic unit of gridding, and integer multiples w are required according to the accuracy of task allocation. m As the grid width, make a preliminary division of the task area.

[0087] Step 2: Assign weights to the importance of each grid in the task area; Assign weights to the importance of the grids in the task area, which is characterized by comprehensively considering factors such as the familiarity F of each area, the current demand and attention A for the area, and the actual number of searches X for the area within a certain period of time, and setting a normalized value K as the importance of the grid.

[0088] The step 2 is specifically as follows:

[0089] S2.1: The number of features required for importance K is UUVs are heading out for exploration;

[0090] in, Indicates rounding up, P is the total number of UUVs,

[0091] The K value is weighted and expressed by the following formula:

[0092]

[0093] Among them, C is a constant based on the average number of searches per unit time in each area;

[0094] S2.2: The F value is calculated using a fuzzy evaluation method. The familiarity of the area is divided into three categories: familiar, general, and unfamiliar, with scores of 0, 0.5, and 1, respectively. The probabilities of familiarity, general, and unfamiliarity in the area are statistically analyzed, the proportions are multiplied by the score weights and normalized, and the sum of the normalized values ​​is taken as the F value. The F value is shown in Table 1. Taking the probability of 10:30:60 as an example in Table 1, F = 0.6 + 0.15 + 0 = 0.75.

[0095] Table 1. F value

[0096] F Familiar(0) Normal (0.5) Unfamiliar(1) Probability (percentage) 10 30 60 Normalization 0 0.15 0.6

[0097] The calculation of A value adopts fuzzy evaluation method, and the evaluation of regional demand and attention is divided into five categories: special attention, attention, some need, dispensable and no attention at all, with scores of 1, 0.75, 0.5, 0.25 and 0 respectively. The probability of each attention is counted, the proportion is multiplied by the score weight and normalized, and the sum of the normalized values ​​is taken as A value; A value is shown in Table 2. For example, Table 2 takes the probability of 50:20:20:10:0 as an example, then A=0.5+0.15+0.1+0.025+0=0.775. ;

[0098] Table 2. A value

[0099]

[0100] According to the statistics of task allocation and execution process, it is the dynamic input quantity X, with the initial value X0=0;

[0101] S2.3: Simplify the process and assign only one K value to each grid area as the importance of the grid. For any grid i, according to its F i , A i , X i The value can be obtained K i , taking three different marks as examples to assign importance to each grid in the region. Figure 2 Taking three different marks as examples, the importance of each grid in the area is weighted.

[0102] In the specific task of searching for static targets on the seabed, the K value is set to dynamically adjust with the number of targets found to achieve multiple detections of key areas, and the average number of targets in each area is taken as T0;

[0103] K T =K+log a (T i -T0)

[0104] The value of a is assigned according to the actual situation, a>1; when T i -T0≤0, take K T =K.

[0105] Step 3: Selection of the largest area within the task area boundary and clustering method of grids with similar importance;

[0106] The step three is specifically as follows:

[0107] S3.1: The method for selecting the largest internal area is to select the coordinate value (x min ,y min )(x max ,y max )(x min ,ymax )(x max ,y min ) of the four grid nodes, the rectangle formed is denoted as Q, and when the number of grids surrounded by Q is the largest, it is denoted as Q max , Q max is the largest internal area;

[0108] S3.2: Cluster and integrate the grids whose importance is within the range of K±I, where I is the inclusiveness, which is between 5%K and 25%K. The difference method is used to test the clustering of the grids with a step size of 1%K and output the results. The I value with reasonable regional division is selected as the specified parameter input value of the algorithm;

[0109] S3.3: Perform artificial secondary aggregation on the parts that can be optimized in the regional division to ensure the reasonable and efficient aggregation of the internal regions;

[0110] S3.4: For areas that are significantly larger than the average partition area (S 聚类 >3S 平均 ) is called the out-of-bounds area. Considering that it takes a long time for a single UUV to search it, it is divided into several areas of appropriate size to improve the search efficiency;

[0111] The importance of the aggregated area is the highest importance value K+I in the area, and the aggregation principle is that low importance can be aggregated to high importance, but not vice versa.

[0112] The principle of dividing the out-of-bounds area into appropriate areas is:

[0113] Segmentation avoids increasing the complexity of search path planning within the area;

[0114] The area of ​​the divided region should be uniform and avoid large difference from the average area;

[0115] Choose a plan with fewer splits.

[0116] To ensure the maximum internal area Q when clustering grids with similar importance max The integrity of the grid is maintained while taking into account the clustering principle of grids with strict similarity in importance; for grids exceeding Q max The boundary grid should follow a stricter rule when clustering, that is, Q max The importance of the grid outside the boundary only falls within the range of K±0.2I, the same as Q max The grids within the boundary are clustered to form erosion and separation of the edge area L.

[0117] Step 4: Refined processing of edge areas;

[0118] The step 4 is specifically as follows:

[0119] S4.1: Q maxThe area enclosed by the boundary extension and the boundary of the task area S is recorded as the edge area L, Q max The area enclosed by the extended edge and the boundary of the task area S starts from the leftmost area, the area with the minimum coordinate value, and is sorted in a clockwise direction, and is recorded as edge areas L1, L2, L3, and L4 respectively. Each area crossing the boundary of the previous area is not counted to avoid overlap;

[0120] S4.2: The edge region L is within the task region SQ max The area outside the max , which is composed of the grids that the boundary of S passes through and its neighboring grids, and is represented by Q max It should be emphasized that since the grids passed by the boundary of S are included, the aggregation of grids in L will form stepped polygons of different shapes.

[0121] S4.3: Clustering of edge region L: The actual effective task area contained in the grid where the boundary passes is different. According to the precision requirement, the grid where the boundary passes is clustered. The standard grid is divided again, and small grids whose intersection with S is empty are removed during clustering;

[0122] S4.4: For the clustering of edge regions L, except for those separated by internal aggregation regions of different importance, all are clustered by S boundary Q max The edge area L formed by the boundary i , aggregated into a stepped polygon or rectangle L i * , for the edge area L separated by the internal areas of different importance i 1 ,L i 2 ,...,L i n Aggregate into edge regions L i *1 , L i *2 ,…,L i *n , where i = 1, 2, 3, 4. Figure 7 Taking L2 as an example, refinement and clustering were performed.

[0123] Through the above steps, the clustering processing of grids of different importance in the task area S can be realized, that is, the division of the task area S is realized. The divided areas are (combined) stepped polygons or rectangles after the adjustment of the inclusiveness I and the manual secondary aggregation processing in the above steps.

[0124] Step 5: Classification of aggregation areas based on geometric features and induction of corresponding search path planning strategies.

[0125] The step five is specifically as follows: Figure 8 Taking as an example, the (geometric) characteristics and search path planning strategies of different types of aggregation areas are summarized as follows: For example, areas 3, 4, 6, and 7 are relatively regular;

[0126] S5.1: For clustering areas that are rectangular or standard stepped polygons, the longest side is used as the travel direction of the UUV comb line scan, and the search is performed from each mission area. After the search is completed, it automatically switches to the longest side direction of the adjacent mission area and performs comb line scan again;

[0127] S5.2: For rectangular or standard stepped polygonal clustering regions with convex or concave grids, where For example, in areas 1 and 5, for the protruding parts, ignore their influence first, search the main area as a regular area according to the strategy of S5.1, and then search the unsearched area; for the concave parts, directly fill them as regular areas, and then search according to the strategy of S5.1;

[0128] S5.3: For cluster regions that are rectangular or standard stepped polygons with convex or concave grids, where The convex or concave grid part is regarded as another adjacent area of ​​the same importance, and the search logic is to search immediately after or before the main area of ​​the rectangular or standard stepped polygonal part; the search strategy for the main area is the same as S5.1, and after the search is completed, it automatically switches to the longest side direction of the convex or concave grid area to perform comb line scanning again, or after searching the convex or concave grid area first, it automatically switches to the longest side direction of the main area to perform comb line scanning. The search order is determined by the distance between the end point of the previous area searched by the UUV and the search starting point of the main area and the convex or concave grid area;

[0129] S5.4: The preset number of detections per unit time in each task area is M ( = represents rounding, P is the total number of UUVs), the initial detection times of each mission area is N = 0, when any UUV searches the designated mission area, the detection times in the area is N = N + 1, and the UUV will be marked; considering the same UUV's re-detection of the mission area, the actual benefit decreases, and the multiple searches of the marked UUV on the same mission area are equivalent to:

[0130] The actual number of detections in the mission area is:

[0131]

[0132] Among them, a iis the judgment function. When the number of detections of the i-th UUV is n i ≥2, a i =1; otherwise a i =0, b i is the judgment function. When the number of detections of the i-th UUV is n i ≥1, b i =1; otherwise b i =0.

[0133] To prevent the situation where all nearby UUVs have completed the search but the number of searches has not reached the importance requirement, and the distant UUVs have to wait too long to perform the search, before the number of searches in the mission area reaches the importance requirement, the neighboring UUVs can be continuously summoned to repeat the search until the importance requirement condition N ≥ M is met.

[0134] The above is a detailed introduction to the task area division and clustering method for collaborative search of multiple underwater robots provided by the present invention. The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A task area division and clustering method for collaborative search of multiple underwater robots, characterized in that: The following steps are involved: Step 1: Preliminary grid division of the mission area based on the UUV operation width; Step 2: Assign weights to the importance of each grid in the task area; Step 3: Selection of the largest area within the task area boundary and clustering method of grids with similar importance; Step 4: Refined processing of edge areas; Step 5: Classification of aggregation areas based on geometric features and induction of corresponding search path planning strategies; The step three is specifically as follows: S3.1: The method for selecting the largest internal area is to select the coordinate value (x min ,y min )(x max ,y max )(x min ,y max )(x max ,y min ) of the four grid nodes, the rectangle formed is denoted as Q, and when the number of grids surrounded by Q is the largest, it is denoted as Q max , Q max is the largest internal area; S3.2: Cluster and integrate the grids whose importance is within the range of K±I, where I is the inclusiveness, which is between 5%K and 25%K. The difference method is used to test the clustering of the grids with a step size of 1%K and output the results. The I value with reasonable regional division is selected as the specified parameter input value of the algorithm; S3.3: Perform artificial secondary aggregation on the parts that can be optimized in the regional division to ensure the reasonable and efficient aggregation of the internal regions; S3.4: For areas S that are significantly larger than the average 聚类 >3S 平均 The clustering area is called the out-of-bounds area. Considering that it takes a long time for a single UUV to search it, it is divided into several areas of appropriate size to improve the search efficiency; The importance of the aggregated region is the highest importance value K+I in the region, and the aggregation principle is that low importance can be aggregated to high importance, but not vice versa; To ensure the maximum internal area Q when clustering grids with similar importance max The integrity of the grid is maintained while taking into account the clustering principle of grids with strict similarity in importance; for grids exceeding Q max The boundary grid should follow a stricter rule when clustering, that is, Q max The importance of the grid outside the boundary only falls within the interval of K±0.2I, and then the same as Q max The grids within the boundary are clustered to form erosion and separation of the edge area L; The step 4 is specifically as follows: S4.1: Q max The area enclosed by the boundary extension and the boundary of the task area S is recorded as the edge area L, Q max The area enclosed by the extended edge and the boundary of the task area S starts from the leftmost area, the area with the minimum coordinate value, and is sorted in a clockwise direction, and is recorded as edge areas L1, L2, L3, and L4 respectively. Each area crossing the boundary of the previous area is not counted to avoid overlap; S4.2: The edge region L is within the task region SQ max Area outside, that is, L = SQ max , which is composed of the grids that the boundary of S passes through and its neighboring grids, and is represented by Q max separated by the borders; S4.3: Clustering of edge region L: The actual effective task area contained in the grid where the boundary passes is different. According to the precision requirement, the grid where the boundary passes is clustered. The standard grid is divided again, and small grids whose intersection with S is empty are removed during clustering; S4.4: For the clustering of edge regions L, except for those separated by internal aggregation regions of different importance, all are clustered by S boundary Q max The edge area L formed by the boundary i , aggregated into a stepped polygon or rectangle L i * , for the edge area L separated by the internal areas of different importance i 1 ,L i 2 ,...,L i n Aggregate into edge regions L i *1 , L i *2 ,…,L i *n , where i=1,2,3,4.

2. The task area division and clustering method for multi-underwater robot collaborative search according to claim 1 is characterized in that: The step 1 is specifically as follows: S1.1: Adaptively configure the unit length of the grid according to the operating width of the UUV performing the task, record UUV1, UUV2, UUV3, ..., UUV n The working width is W1, W2, W3, ..., W n , take the minimum working width as W0: W0=min{w1,w2,...,w n }; S1.2: Order i=1,2,…,n, as the parameter of the actual working width of each UUV compared with the minimum working width; S1.3: Take The lowest common multiple is used as the basic unit of gridding, and integer multiples are required according to the accuracy of task allocation. w m As the grid width, make a preliminary division of the task area.

3. The task area division and clustering method for multi-underwater robot collaborative search according to claim 2 is characterized in that: The step 2 is specifically as follows: S2.1: For the feature representation of importance K, the required number is UUVs are going to explore; in, Indicates rounding up, P is the total number of UUVs, The K value is weighted and expressed by the following formula: Among them, C is a constant based on the average number of searches per unit time in each area; S2.2: The calculation of F value adopts fuzzy evaluation method, and the familiarity of the area is divided into three categories: familiar, general and unfamiliar, with scores of 0, 0.5 and 1 respectively. The probability of familiarity, general and unfamiliarity in the area is counted, and the proportion is multiplied by the score weight and normalized. The sum of the normalized values ​​is taken as the F value; The A value is calculated using a fuzzy evaluation method. The evaluation of regional needs and attention is divided into five categories: special attention, attention, some need, dispensable, and no attention at all, with scores of 1, 0.75, 0.5, 0.25, and 0, respectively. The probability of each attention level is statistically analyzed, and the proportion is multiplied by the score weight and normalized. The sum of the normalized values ​​is taken as the A value. According to the statistics of task allocation and execution process, it is the dynamic input quantity X, with the initial value X0=0; S2.3: Simplify the process and assign only one K value to each grid area as the importance of the grid. For any grid i, according to its F i , A i , X i The value can be obtained K i .

4. The task area division and clustering method for multi-underwater robot collaborative search according to claim 3 is characterized in that: In the specific task of searching for static targets on the seabed, the K value is set to dynamically adjust with the number of targets found to achieve multiple detections of key areas, and the average number of targets in each area is taken as T0; K T =K+log a (T i -T0) The value of a is assigned according to the actual situation, a>1; when T i -T0≤0, take K T =K.

5. The task area division and clustering method for multi-underwater robot collaborative search according to claim 4 is characterized in that: The principle of dividing the out-of-bounds area into appropriate areas is: Segmentation avoids increasing the complexity of search path planning within the area; The area of ​​the divided region should be uniform and avoid large difference from the average area; Choose a plan with fewer splits.

6. The task area division and clustering method for multi-underwater robot collaborative search according to claim 1 is characterized in that: The step five is specifically as follows: S5.1: For clustering areas that are rectangular or standard stepped polygons, the longest side is used as the travel direction of the UUV comb line scan, and the search is performed from each mission area. After the search is completed, it automatically switches to the longest side direction of the adjacent mission area and performs comb line scan again; S5.2: For a rectangular or standard stepped polygonal clustering region with a convex or concave grid, where the convex or concave length For the longest side, ignore the influence of the protruding part first, search the regular area according to the strategy of S5.1, and then search the unsearched area; for the concave part, directly fill it into a regular area, and then search according to the strategy of S5.1; S5.3: For a rectangular or standard stepped polygonal clustering region with a convex or concave grid, where the convex or concave length The longest side, the convex or concave grid part is regarded as another adjacent area of ​​the same importance, and the search logic is to search immediately after or before the main area of ​​the rectangular or standard stepped polygonal part; the search strategy for the main area is the same as S5.1, and after the search is completed, it automatically switches to the longest side direction of the convex or concave grid area and performs comb line scanning again, or after searching the convex or concave grid area first, it automatically switches to the longest side direction of the main area and performs comb line scanning. The search order is determined by the distance between the end point of the previous area searched by the UUV and the search starting point of the main area and the convex or concave grid area; S5.4: The preset number of detections per unit time in each task area is M. represents rounding, P is the total number of UUVs, the initial detection times of each mission area is N=0, when any UUV searches the designated mission area, the detection times in the designated mission area is N+1, and the UUV will be marked; considering the same UUV's re-detection of the mission area, the actual benefit decreases, and the multiple searches of the marked UUV on the same mission area are equivalent to: n≥2, The actual number of detections in the mission area is: Among them, a i is the judgment function. When the number of detections of the i-th UUV is n i ≥2, a i =1; otherwise a i =0, b i is the judgment function. When the number of detections of the i-th UUV is n i ≥1, b i =1; otherwise b i =0.

7. The method for task area division and clustering for collaborative search of multiple underwater robots according to claim 6, characterized in that: To prevent the situation where all nearby UUVs have completed the search but the number of searches has not reached the importance requirement, and the distant UUVs have to wait too long to perform the search, before the number of searches in the mission area reaches the importance requirement, the neighboring UUVs can be continuously summoned to repeat the search until the importance requirement condition N ≥ M is met.

Images