Clustered AUV data-driven submesoscale eddy observation method

Abstract

The application discloses a kind of submesoscale vortex observation methods based on cluster AUV data driving, based on multiple autonomous underwater vehicle, and is observed in water surface and underwater networking cooperation, utilizes AUV high mobility and multi-sensor cooperation, carries CTD, chlorophyll and ADCP and other sensors, to carry out multilayer, multi-profile and section autonomous data acquisition in submesoscale vortex in the mode of formation cooperation;During operation, vortex structure observation data and task information are discussed and shared in each AUV formation and between each formation, based on vortex structure characteristic data, each AUV carries out data-driven online path adjustment between vortex determination layer, section and multilayer, realizes the tracking observation of submesoscale vortex high wisdom, high efficiency, and the scheme can greatly promote the three-dimensional, intelligent and flexible of submesoscale observation, to realize the fine monitoring and identification of submesoscale vortex, evolution process tracking provides advanced technical scheme.

Description

Technical Field

[0001] The present invention belongs to the field of sub-mesoscale eddy observation, and in particular relates to a sub-mesoscale eddy observation method based on cluster AUV data drive. Background Art

[0002] Submesoscale eddies (SMEs) are vortices in the ocean with spatial scales of 1–10 kilometers and time scales of 1–10 days. They play a crucial role in ocean energy cascades and material transport, and have a significant regulatory influence on ocean biogeochemical processes and air-sea interactions. They are a frontier and hot research area in marine science. Currently, there is a lack of means to track and observe the three-dimensional structure and evolution of SMEs, and comprehensive three-dimensional structural data are still lacking.

[0003] The invention patent application with the publication number [CN107655460A] discloses a method for observing mesoscale vortices using underwater gliders. First, an acceleration sensor is installed on the underwater glider to monitor the acceleration and speed of the underwater glider moving with the water body when the underwater glider is in neutral hover. Then, the position, range and movement trend of the mesoscale vortex to be measured are generally judged through the sea surface height anomaly data. Two gliders are used to perform cross-sectional observations of orthogonal paths in the direction of movement of the mesoscale vortex and its normal direction to conduct an initial measurement of the integrity of the mesoscale vortex. Then, four gliders are used to neutrally hover and perform flow observations at the maximum flow velocity zone on the sea surface of the mesoscale vortex, at different depths in the central area, at the maximum flow velocity zone in the maximum jump layer gradient layer, and at the maximum flow velocity zone in the lower uniform layer. In addition, the invention patent with application publication number [CN113741449A] discloses a multi-agent control method for sea-air collaborative observation tasks, which uses a single unmanned boat to search for areas with observation value in the mesoscale vortex; the unmanned boat travels from the outermost side of the vortex in a straight line to the center, and the sensors on the boat collect water temperature at regular intervals, and sort the water temperature data from high to low to obtain areas with large water temperature gradient changes; multiple unmanned boats are dispatched to search for isotherms in the above area, and the navigation attitude of each unmanned boat is continuously controlled using data-driven and deep deterministic policy gradient algorithms to ensure that it travels on the isotherm; a drone is dispatched to the center of the vortex, and a multi-agent deep deterministic policy gradient algorithm is used to control the drone and the unmanned boats to merge.

[0004] However, due to the short life cycle and small spatial scale of submesoscale eddies, the laws of their evolution process are difficult to grasp. Traditional satellite remote sensing, large submersibles / buoys, and ocean surface profile buoy observation methods have low temporal and spatial resolution. Although high-frequency radar (HFR) has high temporal and spatial resolution, it can only observe ocean surface eddies. Although unmanned equipment such as underwater gliders can achieve profile observations, they lack maneuverability and autonomy and cannot track and observe the evolution process of submesoscale eddies.

[0005] Therefore, it is urgent to propose a new observation method suitable for sub-mesoscale eddies, so as to realize the dynamic tracking observation of the three-dimensional structure acquisition and evolution process of sub-mesoscale eddies, promote the understanding and modeling of ocean circulation processes, reveal the deep fine structure and mechanism of sub-mesoscale eddies, especially their dynamic evolution process and other important scientific laws, and provide advanced observation means for my country to carry out high-resolution ocean observations and accumulate rich observation data. Summary of the Invention

[0006] In response to the defects in the existing technology for sub-mesoscale eddy observation, the present invention proposes a sub-mesoscale eddy observation method based on cluster high-intelligence AUV data drive, which combines the determination of levels, sections and multi-level observations to obtain complete sub-mesoscale eddy three-dimensional data.

[0007] The present invention is implemented by adopting the following technical solution: a sub-mesoscale eddy observation method based on cluster AUV data drive, comprising the following steps:

[0008] Step A: Establish an observation area, obtain the position of the submesoscale vortex, and deploy multiple AUVs to the designated area. The AUVs are equipped with multiple different types of sensors, including but not limited to CTD, chlorophyll, and ADCP;

[0009] Step B: Forming a formation of multiple AUVs, and tracking and observing the sub-mesoscale vortex at a certain level based on the sub-mesoscale vortex characteristic distribution density and vortex boundary characteristic identification results;

[0010] Step C: planning the AUV formation cross-sectional observation path according to the movement direction of the submesoscale vortex to achieve cross-sectional observation;

[0011] Step D: preset the observation path for the next level based on the vortex center position and vortex motion trajectory, and observe the next level according to the evolution direction of the sub-mesoscale vortex using the method of step B;

[0012] Step E: Repeat steps B to D until the sub-mesoscale eddy observation is completed.

[0013] Furthermore, in step B, when performing tracking observation at a certain level, the following steps are specifically included:

[0014] Step B1: Establish a Gaussian regression model based on the observed data:

[0015] (1) Data acquisition: Obtain the online recognition results of the boundary feature data of the submesoscale vortex and process the time series of the recognition results based on the sliding average filter method;

[0016] (2) Prediction model update: Combine the recognition results of multiple AUVs sharing their own feature data and update the Gaussian regression model according to the recognition results;

[0017] Step B2: Observation path planning at a certain level: predict the data of the unobserved area based on the Gaussian regression model, calculate the regional gradient extreme value, select the observation direction based on the task mode and gradient extreme value, and finally determine the task execution status.

[0018] Furthermore, in step B1, the following method is specifically adopted:

[0019] (1) Each AUV synchronizes and correlates multiple sensor time series signals (MTS) acquired in real time. Using the anomaly threshold range obtained from numerical simulation as prior knowledge, the preprocessed MTS is modeled for local and global features to obtain reconstructed time series information. Anomalies are detected based on the reconstruction error, enabling online identification of eddy boundaries.

[0020] (2) After completing the vortex boundary feature data recognition within a certain time interval, the AUV transmits the recognition results to other members in the formation; based on the recognition results of its own accumulated observation data and the information shared within the formation, each AUV updates the Gaussian process regression (GPR) model according to the recognition results.

[0021] Furthermore, in step B1, the online recognition result of the eddy boundary feature data obtained by the AUV is represented by 0 or 1, 0 represents not a boundary, and 1 represents a boundary; the recognition result time series is processed based on the sliding average filtering method, and n recognition results are stored in sequence. When a new recognition result is obtained, the earliest received result is discarded, and then the arithmetic mean of the n recognition results including the new result is solved. The average value of the filtered boundary recognition result obtained at the corresponding sampling point is recorded as Y, Y = {y1, y2, ..., y n}, i = 1 to n, y i Represents the i-th boundary recognition result.

[0022] Furthermore, in step B2, the following method is specifically adopted:

[0023] (1) The AUV determines which observation mode it is currently in based on the average value Y of the boundary recognition result. When the AUV is in mode A or mode B, it performs sampling according to the preset route. Mode A refers to sampling observation inside the vortex, and mode B refers to sampling observation at the vortex boundary and outside.

[0024] (2) When the AUV is in mode A and switches to mode B, the motion range of the AUV before the next sampling point is limited to the current position P of the AUV. i As the origin, R s In the semicircular area with radius R s =T s ·v s , T s is the time interval for obtaining recognition results, vs is the speed of the AUV; calculate From P i Move to P in the direction of increasing gradient i+1 , the estimate of the observed value obtained from the Gaussian regression model Calculate the estimated gradient value to ensure that the gradient is maximum and the direction is positive. The position corresponding to the maximum gradient value of the predicted observation value is the planned path point.

[0025] (3) When the AUV switches from mode B to mode A, the main observation task is to search and observe at the boundary and outside of the vortex. At this time, the preset path is used as a constraint, data sampling is performed on the outside of the vortex, and after completing the data sampling, it returns to the inside of the vortex as soon as possible to continue the observation task. At this time, the position corresponding to the minimum gradient value of the predicted observation value is the planned path point. When the path point exceeds the preset path range, the AUV is forced to return to the preset path.

[0026] Furthermore, in step D, after completing the cross-sectional observation, the AUV obtains the vortex center position of the previous level based on the recognition result of the multi-dimensional sensor fusion data, and uses a fast search based on the density peak clustering algorithm to extract the shape of the sub-mesoscale vortex of the previous observation level;

[0027] Combining the vortex center position and vortex shape, multiple AUVs predict the structural information of the vortex at the next level to be observed, generate the minimum circumscribed rectangle of the vortex as the initial observation area, and generate a preset observation path according to the method in step B to conduct observations at the next level.

[0028] Furthermore, in step B2, when the AUV is in mode switching, the change trend of the element in Y is used to determine whether the current mode is switching from mode A to mode B or from mode B to mode A. The determination method is as follows:

[0029]

[0030] Among them, y i Indicates the i-th boundary recognition result, y i-1 represents the i-1th boundary recognition result, and n represents the number of boundaries.

[0031] Furthermore, in step B, during the AUV observation process, after each path planning, it is necessary to determine whether the current observation is in mode A or mode B; repeat steps B1-B2 until the cluster AUV completes the observation of the currently determined level of the sub-mesoscale vortex.

[0032] Furthermore, in step B2:

[0033] The maximum gradient is calculated as follows:

[0034]

[0035] The minimum gradient calculation method is as follows:

[0036]

[0037] in, represents the boundary recognition result at the i+1th path point based on Gaussian process estimation, △P represents the change in the AUV position, which can be approximated by the constant R when the speed is determined. s , max(*) and min(*) represent the maximum and minimum values ​​of * respectively, represents the gradient estimate.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are:

[0039] This scheme is based on the use of multiple AUVs in a coordinated formation to track and observe the three-dimensional structure and evolution of sub-mesoscale vortices. The AUVs are equipped with multiple sensors such as CTD, chlorophyll, and ADCP. Based on data-driven tracking and observation of sub-mesoscale vortices at a certain level, the AUV formation cross-sectional observation path is then planned. Based on the evolution direction of the sub-mesoscale vortex, the path planning for the next level is carried out. Combining the observations at the certain level, cross-section, and multiple levels, the overall tracking and observation of the vortex is achieved, and the three-dimensional structure of the vortex is obtained.

[0040] This plan can greatly promote the three-dimensional, automated, intelligent and flexible observation of sub-mesoscale processes, provide advanced technical solutions for the refined monitoring and identification of sub-mesoscale eddies, tracking of evolution processes, and tracking observations, and realize the typical application of AUV-based three-dimensional structure and tracking observations of ocean sub-mesoscale eddies, revealing important marine scientific laws such as the evolution mechanism of sub-mesoscale dynamic processes and nutrient transport processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the process of observing sub-mesoscale eddies according to an embodiment of the present invention;

[0042] Figure 2 Schematic diagram of the observation path at the submesoscale eddy level, (a) is the preset path; (b) is the actual path based on data-driven planning;

[0043] Figure 3 A flowchart of a data-driven observation method at a determination level according to an embodiment of the present invention is provided;

[0044] Figure 4 Schematic diagram of an observation method for switching from mode A to mode B according to an embodiment of the present invention;

[0045] Figure 5 Schematic diagram of the observation method for switching from mode B to mode A in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described below with reference to the accompanying drawings and embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0047] The solution of the present invention is based on multiple highly intelligent autonomous underwater vehicles (AUVs) for surface and underwater networked collaborative observation. The AUVs are equipped with a variety of different types of sensors, including CTD, chlorophyll, and ADCP. Utilizing the high maneuverability and multi-sensor collaboration of the AUVs, autonomous data collection is performed on multiple levels, multiple profiles, and sections in sub-mesoscale vortices in a formation-coordinated mode. During the operation, vortex structure observation data and mission information are discussed and shared within and between each AUV formation. Based on the vortex structure characteristic data, each AUV performs data-driven online path adjustment at the vortex's determined level, section, and multiple levels, achieving highly intelligent and efficient tracking and observation of sub-mesoscale vortices.

[0048] Specifically, such as Figure 1 As shown, this embodiment proposes a sub-mesoscale eddy observation method based on cluster high-intelligence AUV data drive, which uses multiple AUVs to accurately observe the sub-mesoscale eddy layer and cross-section to collect rich multi-dimensional observation data, including the following steps:

[0049] The method involves establishing an observation area, obtaining the location of submesoscale eddies, and deploying multiple AUVs equipped with various sensors, including CTD, chlorophyll, and ADCP, to the designated area. The method involves forming a formation of multiple AUVs and tracking and observing the submesoscale eddies at a specific level based on the density of submesoscale eddy distribution characteristics and eddy boundary feature identification results.

[0050] Step C: Conduct cross-sectional observation according to the sub-mesoscale eddy observation area to realize the cross-sectional observation path planning research of the AUV formation;

[0051] Step D: preset the observation path for the next level based on the vortex center position and vortex motion trajectory, and use the method of step B to achieve the next level of observation according to the evolution direction of the sub-mesoscale vortex;

[0052] Step E: Repeat steps B to D until the sub-mesoscale eddy observation is completed.

[0053] Specifically, the embodiments of the present invention are described in detail below:

[0054] First, establish a preliminary observation area and observe submesoscale eddies on the sea surface based on satellite remote sensing data, numerical simulation results, and drone aerial photography. Then, multiple AUVs are deployed to the designated area via a mother ship. This example uses the deployment of 10 AUVs as an example.

[0055] 2. Conduct data-driven tracking observations of submesoscale eddies at a certain level;

[0056] In this embodiment, 5 AUVs are divided into two formations, namely Figure 1 In formation 1 and formation 2, the AUV first observes at a certain level of the submesoscale eddy. The observation plan at the determined level is as follows: Figure 3 As shown, the current path planning method for determining the level includes the following steps:

[0057] Step 21: Environmental modeling based on observation data:

[0058] (1) Data acquisition and utilization: Obtain online recognition results of eddy boundary feature data and process the recognition result time series based on the sliding average filter method;

[0059] As attached Figure 2 As shown in (a), a bounding rectangle of the vortex (sub-mesoscale vortex) to be observed is drawn to generate a preset path similar to the fixed-line observation route in traditional ocean observations; two formations perform sensor data sampling operations along the preset route; each AUV synchronizes and correlates the multi-sensors time series (MTS) signals acquired in real time, using the anomaly threshold range obtained from numerical simulation as prior knowledge. The pre-processed MTS is modeled for local and global features to obtain richer reconstructed time series information, and anomalies are detected based on the reconstruction error, achieving accurate online identification of vortex boundaries;

[0060] Among them, the online recognition result of the boundary feature data obtained by the AUV is represented by 0 or 1, 0 represents not a boundary, and 1 represents a boundary. The recognition result time series is processed based on the sliding average filter method, and n recognition results are stored in order. Every time a new recognition result is obtained, the earliest result received is discarded, and then the arithmetic mean of the n recognition results including the new result is calculated. In this way, a new average value can be calculated for each recognition result received. The average value of the filtered boundary recognition result obtained at the corresponding sampling point is recorded as Y, Y={y1,y2,…,y n}, i=1~n.

[0061] (2) Prediction model update: Combine the recognition results of multiple AUVs sharing their own feature data and update the Gaussian regression model based on the recognition results:

[0062] Based on observations along a preset path, vortex structure observation data and mission information are shared within and between AUV formations. After identifying vortex boundary feature data within a certain time interval, the AUVs broadcast the identification results to other members of the formation using underwater acoustic communication or 4 / 5G communication. Based on the identification results of their own accumulated observation data and the information shared within the formation, each AUV updates the Gaussian process regression (GPR) model based on the identification results, and uses GPR analysis to predict environmental data in unobserved areas. Based on the prediction and identification results, the AUVs update the current environmental model in real time and predict the observation data of unsampled points based on the estimated model. As the observation progresses, new observation points are also used as training samples to update the Gaussian regression model.

[0063] Step 22: Observation path planning for a certain level: predict data of unobserved areas based on GPR, calculate regional gradient extremes, select observation directions based on mission mode and gradient extremes, and finally determine mission execution status;

[0064] from Figure 2 (a) It can also be seen that the AUV is not allowed to scan repeatedly on the same path based on the preset path constraint, so the AUV's forward direction is set to turn 90 degrees left to turn 90 degrees right. This embodiment divides the data-driven path planning at the determination level into two modes: vortex internal sampling observation (mode A) and vortex boundary and external sampling observation (mode B). The AUV determines which observation mode it is currently in based on Y, and the judgment conditions are as follows:

[0065]

[0066] (1) When the AUV is in mode A or mode B, sampling is performed according to the preset route;

[0067] (2) When the AUV is in mode switching, the change trend of the elements in Y is used to determine whether the current mode is switching from mode A to mode B or vice versa. The judgment method is as follows:

[0068]

[0069] 1) When the AUV is in mode A->mode B, such as Figure 4 As shown, the motion range of the AUV before the next sampling point is limited to the current position of the AUV P i As the origin, R s is the radius, R s =T s ·v s In the semicircular area, T s is the time interval for obtaining recognition results, v s is the speed of the AUV. Calculate From P iMove to P in the direction of increasing gradient i+1 Because △P=||P i+1 -P i ||=R s is a constant value, so the estimated value of the observation obtained by GPR can be Calculate the estimated gradient to ensure that the gradient is maximum and the direction is positive. The location corresponding to the maximum value of the predicted observation value is the planned path point. The maximum gradient calculation method is as follows:

[0070]

[0071] Determine whether the re-planned path point is within the range specified by the preset path. If it exceeds, the re-planned path point is restricted to the preset path; if it does not exceed, the re-planned path point is used as the next data sampling point of the AUV;

[0072] 2) When the AUV is in mode B->mode A, such as Figure 5 As shown in the figure, the main observation task is to search and observe at the boundary and outside of the vortex. At this time, data sampling is performed on the outside of the vortex with the preset path as the constraint, and after completing the data sampling, the AUV returns to the inside of the vortex as soon as possible to continue the observation task. At this time, the position corresponding to the minimum gradient value (negative value) of the predicted observation value is the planned path point. When the path point exceeds the preset path range, the AUV is forced to return to the preset path. The minimum gradient calculation method is as follows:

[0073]

[0074] During the AUV observation process, after each path planning, it is necessary to determine whether the current observation is in mode A or mode B, so as to realize online path planning based on the eddy characteristic data. The re-planned path is as follows: Figure 2 As shown in (b), sufficient data volume is ensured and observation efficiency is improved; steps 21-22 are repeated until the cluster AUVs complete the observation of the currently determined level of the sub-mesoscale vortex.

[0075] 3. Conduct research on cross-sectional observation path planning for AUV formations; conduct cross-sectional observations using multiple AUVs, and plan a cross-sectional observation path that conforms to the direction of vortex motion and has optimal energy consumption in combination with the evolution direction of sub-mesoscale eddies.

[0076] 4. According to the evolution direction of the sub-mesoscale eddy, plan the next level path to achieve overall tracking and observation of the eddy and obtain the three-dimensional structure of the eddy.

[0077] After completing the cross-sectional observation, the AUVs determine the vortex center location of the previous level based on the recognition results of the multi-dimensional sensor fusion data. Using the fast search based on density peak clustering (CFSFDP) algorithm, they extract the shape of the sub-mesoscale vortex at the previous level. Combining the vortex center location and vortex shape, multiple AUVs predict the structure of the vortex at the next level to be observed, generating the minimum bounding rectangle of the vortex as the initial observation area. Following the method described in step two, they generate a pre-set observation path for observation at the next level.

[0078] 5. Repeat steps 2 to 4 until submesoscale eddy detection is completed.

[0079] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any person skilled in the art may utilize the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes for application in other fields. However, any simple modification, equivalent change, and modification of the above embodiments made in accordance with the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A sub-mesoscale eddy observation method based on cluster AUV data drive, characterized by: The following steps are involved: Step A: Establish an observation area, obtain the position of the submesoscale vortex, and deploy multiple AUVs to the designated area. The AUVs are equipped with multiple different types of sensors, including but not limited to CTD, chlorophyll, and ADCP; Step B: Forming a formation of multiple AUVs, and tracking and observing the sub-mesoscale vortex at a certain level based on the sub-mesoscale vortex characteristic distribution density and vortex boundary characteristic identification results; When conducting tracking observations at a certain level, the following steps are specifically included: Step B1: Establish a Gaussian regression model based on the observed data: (1) Data acquisition: Obtain online recognition results of submesoscale eddy boundary feature data and process the recognition result time series based on the sliding average filter method; (2) Prediction model update: Combine the recognition results of multiple AUVs sharing their own feature data and update the Gaussian regression model based on the recognition results; Step B2: Observation path planning at a certain level: predict data in unobserved areas based on the Gaussian regression model, calculate regional gradient extremes, select observation directions based on the task mode and gradient extremes, and ultimately determine the task execution status; (1) AUV based on the average value of boundary recognition results Determine which observation mode the AUV is currently in. When the AUV is in mode A or mode B, it performs sampling along the preset route. Mode A refers to sampling observations inside the vortex, while mode B refers to sampling observations at the vortex boundary and outside. (2) When the AUV is in mode A and switches to mode B, the movement range of the AUV before the next sampling point is limited to the current position of the AUV. As the origin, In the semicircular area with a radius of , is the time interval for obtaining recognition results, is the speed of the AUV; calculate from Move in the direction of increasing gradient , the estimate of the observed value obtained from the Gaussian regression model Calculate the estimated gradient value to ensure that the gradient is maximum and the direction is positive. The position corresponding to the maximum gradient value of the predicted observation value is the planned path point. (3) When the AUV switches from mode B to mode A, the main observation task is to search and observe at the boundary and outside of the vortex. At this time, the preset path is used as a constraint to perform data sampling outside the vortex, and after completing the data sampling, it returns to the inside of the vortex as soon as possible to continue the observation task. At this time, the position corresponding to the minimum gradient value of the predicted observation value is the planned path point. When the path point exceeds the preset path range, the AUV is forced to return to the preset path; Step C: planning the AUV formation cross-sectional observation path according to the movement direction of the submesoscale vortex to achieve cross-sectional observation; Step D: preset the observation path for the next level based on the vortex center position and vortex motion trajectory, and observe the next level according to the evolution direction of the sub-mesoscale vortex using the method of step B; Step E: Repeat steps B to D until the sub-mesoscale eddy observation is completed.

2. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 1 is characterized in that: In step B1, the following method is specifically adopted: (1) Each AUV synchronizes and correlates the MTS signals of multiple sensors acquired in real time. The anomaly threshold range obtained from numerical simulation is used as prior knowledge. The local and global features of the pre-processed MTS are modeled to obtain reconstructed time series information. Anomalies are detected based on the reconstruction error, and online identification of eddy boundaries is achieved. (2) After completing the identification of vortex boundary feature data within a certain time interval, the AUV transmits the identification results to other members in the formation; based on the identification results of its own accumulated observation data and the information shared within the formation, each AUV updates the Gaussian process regression (GPR) model according to the identification results.

3. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 2 is characterized in that: In step B1, the online recognition result of the vortex boundary characteristic data obtained by the AUV is represented by 0 or 1, where 0 represents not being a boundary and 1 represents being a boundary; Based on the sliding average filter method, the recognition result time series is processed. The n recognition results are stored in order. When a new recognition result is obtained, the earliest result is discarded. Then the arithmetic mean of the n recognition results including the new result is calculated. The average value of the filtered boundary recognition result obtained at the corresponding sampling point is recorded as , , , Indicates the Boundary recognition results.

4. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 1, characterized in that: In step D, after completing the cross-sectional observation, the AUV obtains the vortex center position of the previous level based on the recognition result of the multi-dimensional sensor fusion data, and uses a fast search based on the density peak clustering algorithm to extract the shape of the sub-mesoscale vortex of the previous observation level; Combining the vortex center position and vortex shape, multiple AUVs predict the structural information of the vortex at the next level to be observed, generate the minimum circumscribed rectangle of the vortex as the initial observation area, and generate a preset observation path according to the method in step B to conduct observations at the next level.

5. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 1, characterized in that: In step B2, when the AUV is in mode switching, The changing trend of the elements in the image is used to determine whether the current mode is switched from mode A to mode B or from mode B to mode A. The judgment method is as follows: ； in, Indicates the Boundary recognition results, Indicates the boundary recognition results, and n represents the number of boundaries.

6. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 1, characterized in that: In step B, during the AUV observation process, after each path planning, it is necessary to determine whether the current observation is in mode A or mode B; repeat steps B1-B2 until the cluster AUV completes the observation of the currently determined level of the sub-mesoscale vortex.

7. The sub-mesoscale eddy observation method based on cluster AUV data drive according to claim 1, characterized in that: In step B2: The maximum gradient is calculated as follows: ； The minimum gradient calculation method is as follows: ； in, Represents the first The boundary recognition results at the path points are: Indicates the change in the AUV position, which can be approximated to a constant when the speed is determined , and Respectively The maximum and minimum values ​​of represents the gradient estimate.

Images