Multi-AUV (Autonomous Underwater Vehicle) collaborative deep sea cable searching method based on neural network
By using neural network models in multi-AUV collaborative deep-sea cable search, the effective search of AUV in the subsea task area is guided, and the problem of lack of collaborative exploration of AUV collaborative search in the existing technology is solved, and the precise detection and regional coverage of subsea cables are achieved.
Patent Information
- Application Number
- CN202410934600.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-22
AI Technical Summary
After the existing multi-AUV collaborative deep-sea cable search method, there is a lack of collaborative exploration between AUVs, and the segmentation method is complex and difficult to implement, making it difficult to achieve full coverage search of the subsea area.
Using a multi-AUV collaborative deep-sea cable search method based on neural networks, a two-dimensional neural network model is generated by establishing communication connections between AUVs, and a two-dimensional neural network model is generated by Glasius bionic neural network, and a comprehensive score is determined based on the grid map and neuronal activity values to guide AUV in Effective search and obstacle avoidance of subsea mission areas to ensure full coverage of the area.
It realizes effective coordinated search of AUV in the subsea mission area, avoids collisions and repeated searches, improves search efficiency, and ensures accurate detection of subsea cables and complete coverage of the area.
Smart Images

Figure CN119005268A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of deep-sea target detection, and in particular to a multi-AUV collaborative deep-sea cable search method based on a neural network. Background Art
[0002] The deep-sea cable collaborative search mission refers to the full coverage of an unknown area on the seabed to obtain the specific location of the cable target, which can provide information support for subsequent target strike missions. In order to search for cable targets efficiently and accurately, multiple autonomous underwater vehicles (AUVs) are required to collaboratively detect, avoid inaccessible areas, cover the entire unknown area, and find the location of all target cables. Due to the complex and changeable seabed environment and the need to collaboratively control multiple AUVs, the multi-AUV underwater collaborative search mission has become a very challenging research field.
[0003] Underwater collaborative search based on AUVs has become a current research hotspot. Maza and Ollero proposed a solution that uses polygonal area segmentation and effective coverage algorithm to perform multi-AUV search tasks. Some scholars also proposed a solution that uses Voronoi diagrams to segment the search area to perform multi-AUV search tasks. The core idea of these solutions is to divide the search area into multiple sub-areas, thereby simplifying the multi-AUV collaborative search problem into a single-machine coverage search problem in each sub-area. However, in this approach, after the search area is divided, different AUVs do not actually achieve collaborative exploration, and these area segmentation methods are very complex and uncertain, making them difficult to implement in engineering. Summary of the invention
[0004] In response to the above problems and technical requirements, this application proposes a multi-AUV collaborative deep-sea cable search method based on a neural network. The technical solution of this application is as follows:
[0005] A multi-AUV collaborative deep-sea cable search method based on a neural network comprises the following steps:
[0006] N AUVs establish communication connections between each other, and the integer parameter N≥2. The multi-AUV collaborative deep-sea cable search method includes: i :
[0007] The seabed mission area is rasterized to obtain a grid map, and the Glasius bionic neural network is used to generate a two-dimensional neural network model based on the grid map. The neurons in the two-dimensional neural network model correspond one-to-one to the grids of the grid map, and there is a connection relationship between the neurons corresponding to the two grids within the single-moment motion step of each AUV.
[0008] At time t=0, the activity value of each neuron in the two-dimensional neural network model is initialized according to the grid state of each grid, and the position information of the current grid is initialized; the grid state of each grid is any one of the forbidden state, the searched state and the unsearched state;
[0009] At any time t, the position information of the grids where other AUVs are located is obtained through the communication connection with other AUVs; for any neighboring grid g that is not searched within the single-time motion step of the AUV in the current grid, according to the AUV i Its own position information and the position information of other AUVs are used to evaluate the AUV when it moves to the neighborhood grid g. i The comprehensive dispersion index f between each other AUV g (t), and according to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t) Combined with the comprehensive dispersion index f g (t) Determine the comprehensive score F of the neighborhood grid g g (t), dispersion index f g (t) indicates that the higher the distribution dispersion between the AUVs and other AUVs, the higher the activity value of the neuron and the higher the comprehensive score;
[0010] At time t+1, it moves to the neighborhood grid with the highest comprehensive score for detection, updates the grid status of each grid, and updates the activity value of each neuron at time t+1 based on the grid status of each grid and the connection relationship of each neuron until the search of the seabed mission area is completed.
[0011] A further technical solution is to determine the comprehensive score of the neighborhood grid g including:
[0012] According to AUV i Its own position information and any AUV j The position information of the AUV is determined i When moving to the neighborhood grid g, the AUV j The distance cost J d (t) and relative position And determine the AUV i and AUV j The dispersion index m d is the coefficient; where AUV i When moving to the neighborhood grid g, the AUV j The closer the distance, the greater the distance cost J d (t) The larger the AUV i With AUV j The greater the difference in direction angles, the greater the relative orientation The larger the value, the integer parameter j∈[1,N] and j≠i;
[0013] According to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t), combined with other AUVs j The dispersion index f g j (t) Determine the comprehensive score F of the neighborhood grid g g (t).
[0014] Its further technical solution is to determine the AUV i When moving to the neighborhood grid g, the AUV j Relative position of include:
[0015] According to AUV i The AUV is determined by its own position information and the relative position relationship between the neighborhood grid g and the grid at the current time t. i The heading angle of the movement to the neighborhood grid g
[0016] According to AUV i Own location information and AUV j The position information of the AUV is determined i At the current time t, with AUV j The relative direction between
[0017] According to the heading angle and relative direction Identify AUV i When moving to the neighborhood grid g, the AUV j Relative position of
[0018] Its further technical solution is that AUV i When moving to the neighborhood grid g, the AUV j The distance cost d ij (t) is AUV i When moving to the neighborhood grid g, the AUV j The distance between s is the predetermined minimum distance threshold.
[0019] Its further technical solution is to determine the AUV i When moving to the neighborhood grid g, the AUV j Relative position of
[0020]
[0021] Its further technical solution is to determine the AUV i The heading angle of the movement to the neighborhood grid g Includes: According to AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and the location coordinates of the neighborhood grid g (x g ,y g ), determine the AUV i The heading angle of the movement to the neighborhood grid g
[0022] Its further technical solution is to determine the AUV i At the current time t, with AUV j The relative direction between Includes: According to AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and AUV j The position coordinates of the grid at the current time t (l xj (t),l yj (t)), determine the AUV i With AUV j The relative direction between
[0023]
[0024] A further technical solution is to determine the comprehensive score F of the neighborhood grid g g (t) = z g (t)+f g (t), where the comprehensive dispersion index
[0025] A further technical solution is to update the activity value of each neuron at time t+1 according to the grid state of each grid and the connection relationship of each neuron, including for any neuron m:
[0026] According to the grid state of the grid corresponding to neuron m, the external input signal I of neuron m is determined m ; When the grid state of the grid corresponding to neuron m is unsearched, I m =E; when the grid state of the grid corresponding to neuron m is forbidden, I m = -E; when the grid state of the grid corresponding to neuron m is the searched state, I m =0; E is a positive constant much greater than 1;
[0027] According to the activity value z of any neuron n connected to neuron m at time t n (t), determine the activity value of neuron m at time t+1 Among them, [z n (t)] + =max(z n (t),0), N(m) is the set of neurons connected to neuron m, W mn is the connection weight between neuron m and neuron n and W mn =W nm ; Parameter β>0.
[0028] A further technical solution is that the connection weight W between neuron m and neuron n is mn for:
[0029]
[0030] Where γ is a positive constant coefficient and γ∈(0,1], |mn| is the distance between neuron m and neuron n, and R is the range radius of neuron connection.
[0031] The beneficial technical effects of this application are:
[0032] This application proposes a neural network-based multi-AUV collaborative deep-sea cable search method, which uses the Glasius bionic neural network to simulate the seabed mission area, and realizes the mapping of the AUV's motion mode and the neural network model in the real seabed mission area scene. The neuron activity value is used to represent the state of the corresponding seabed area, guiding the AUV to effectively avoid obstacles and move towards the unsearched area, ensuring the complete search coverage of the entire seabed mission area, which is conducive to the accurate detection of submarine cable targets.
[0033] By establishing communication connections between multiple AUVs to obtain each other's location information, the movement path of each AUV is obtained based on the comprehensive discreteness of multiple AUVs in the seabed mission area and the neuron activity value. This helps to use the status of each AUV and the overall coverage of the seabed mission area to adjust the search path in real time, control the collaborative search between multiple AUVs, effectively improve the efficiency of the search task, and achieve rapid detection of submarine cables.
[0034] The distribution between any two AUVs is determined by calculating the distance and relative orientation between them, and then the comprehensive discreteness of multiple AUVs in the seabed mission area is determined. It can ensure that when multiple AUVs are searching collaboratively, each AUV will preferentially move in the opposite direction to each other and maintain the distance between AUVs to prevent collisions and repeated searches. It ensures the safety and effectiveness of the collaborative search for submarine cables by multiple AUVs, which is of great significance to subsequent target strike missions. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 This is a diagram of the multi-AUV distributed communication architecture.
[0036] Figure 2 It is a flow chart of the multi-AUV collaborative deep-sea cable search method.
[0037] Figure 3 This is a schematic diagram of the correspondence between neurons and grids.
[0038] Figure 4 It is a schematic diagram of the coordinated motion of multiple AUVs.
[0039] Figure 5 It is the coverage curve of the seabed mission area search.
[0040] Figure 6 It is a coverage map of the seabed mission area.
[0041] Figure 7 This is the neuron map corresponding to the seabed mission area.
[0042] Figure 8 This is a comparison chart of the search effects with different numbers of AUVs. DETAILED DESCRIPTION
[0043] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.
[0044] The present application discloses a multi-AUV collaborative deep-sea cable search method based on a neural network. Figure 1 The multi-AUV distributed communication architecture shown in the figure is implemented, and the multi-AUV distributed communication architecture includes N AUVs, and communication connections are established between any two AUVs, where N is an integer and N≥2, such as Figure 1The multi-AUV distributed communication architecture includes three AUVs, namely AUV1, AUV2, and AUV3, for example. Any AUV in the multi-AUV distributed communication architecture executes the multi-AUV collaborative deep-sea cable search method of this application to autonomously calculate the path, which can give full play to the autonomous capabilities of each AUV. Although the centralized communication architecture can coordinate and optimize the search problem from a global perspective, its central node is vulnerable in practical applications. Therefore, the system robustness of the multi-AUV distributed communication architecture adopted in this application is stronger than that of the centralized communication architecture, and is more suitable for deep-sea scenarios with higher reliability requirements.
[0045] Please refer to Figure 2 As shown in the flowchart, the multi-AUV collaborative deep-sea cable search method is for any AUV i Follow these steps:
[0046] Step 1: rasterize the seabed mission area to obtain a grid map, and use the Glasius bionic neural network to generate a two-dimensional neural network model based on the grid map. The neurons in the two-dimensional neural network model correspond one-to-one to the grids of the grid map, and there is a connection relationship between the neurons corresponding to the two grids within the single-moment motion step of each AUV.
[0047] There are deep-sea cables in the unknown seabed mission area. The multi-AUV collaborative search task is to conduct an area search on the seabed mission area to obtain the specific location of the deep-sea cables. After the AUV enters the mission area, it uses a sensor to detect at a certain height from the seabed, and regards the movement of the AUV as a two-dimensional space movement. For a seabed mission area of any shape, in one embodiment, the seabed mission area is simplified using an external rectangle, and the rectangular area is divided into discrete grids of the same size to obtain a grid map. The position of each discrete grid is represented by a two-dimensional coordinate, and the length and width of the discrete grid are set to the navigation distance of the AUV at an average speed in one time step. In each time step, the AUV detects the current grid and can move to any other grid within the eight neighborhoods of the current grid in the next time step. This application defines that each grid and any other grid within its eight neighborhoods can access each other.
[0048] In one embodiment, a fully connected two-dimensional Glasius bionic neural network is used to represent the gridded seafloor mission area, and each neuron is connected to its surrounding neurons, such as Figure 3 The neurons j, g, and k shown in (a) correspond to Figure 3 (b) Grids j, g, and k are shown. By corresponding each neuron to each grid in the grid map, connections are established between neurons for grids that can access each other, and a two-dimensional neural network model is obtained.
[0049] Step 2, at time t = 0, initialize the activity value of each neuron in the two-dimensional neural network model according to the grid state of each grid, and initialize the location information of the current grid. Due to the complex seabed terrain, there are some areas that are prohibited from entering. During initialization, it is necessary to initialize the grid state of the grid map corresponding to the seabed mission area, and initialize the activity value of each neuron in the two-dimensional neural network model according to the grid state of each grid. The grid state of each grid is any one of the forbidden state, searched state and unsearched state. The grid state in the forbidden entry area remains in the forbidden state, indicating that the AUV is prohibited from entering the area. The grid state in the area other than the forbidden entry area is the searched state and the unsearched state. The searched state indicates that an AUV has arrived at the grid and conducted detection, and the unsearched state indicates that no AUV has arrived at the grid and conducted detection.
[0050] Step 3: At any time t, the position information of the grids where other AUVs are located is obtained through the communication connection with other AUVs. For any neighboring grid g that is in the unsearched state within the single-time motion step of the AUV in the current grid, according to the AUV i Its own position information and the position information of other AUVs are used to evaluate the AUV when it moves to the neighborhood grid g. i The comprehensive dispersion index f between each other AUV g (t), and according to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t) Combined with the comprehensive dispersion index f g (t) Determine the comprehensive score F of the neighborhood grid g g (t), dispersion index f g (t) indicates that the higher the distribution discreteness between the AUV and other AUVs, the higher the activity value of the neuron and the higher the comprehensive score. For the grids that are in the forbidden state or searched state within the single-moment movement step of the AUV in the current grid, the method of the present application excludes the grid from the target range when selecting the next moving grid, and no longer calculates the comprehensive score of the grid, which can improve the operation efficiency and reduce redundant calculations.
[0051] Step 4: At time t+1, move to the neighborhood grid with the highest comprehensive score for detection, update the grid status of each grid, and update the activity value of each neuron at time t+1 according to the grid status of each grid and the connection relationship of each neuron, until the search of the submarine mission area is completed and the target position of the submarine cable is obtained.
[0052] The activity value of a neuron is transmitted through the connection relationship between neurons, and the connection relationship corresponds to the grid state of the neighborhood grid. In one embodiment, the external input signal I of neuron m is determined based on the grid state of the grid corresponding to neuron m. m ; When the grid state of the grid corresponding to neuron m is unsearched, I m =E; when the grid state of the grid corresponding to neuron m is forbidden, I m = -E; when the grid state of the grid corresponding to neuron m is the searched state, I m =0; E is a normal number much greater than 1.
[0053] According to the activity value z of any neuron n connected to neuron m at time t n (t), determine the activity value of neuron m at time t+1 Among them, [z n (t)] + =max(z n (t),0), indicating that the positive activity value has a global propagation effect and can be propagated outward through the connection relationship between neurons, while the negative activity value of the neuron has a local effect and will not propagate outward. N(m) is the set of neurons connected to neuron m, W mn is the connection weight between neuron m and neuron n and W mn =W nm , Parameter β>0. The connection weight W between neuron m and neuron n mn for:
[0054]
[0055] Where γ is a positive constant coefficient and γ∈(0,1], |mn| is the distance between neuron m and neuron n, and R is the range radius of neuron connection. The distance between neurons will be connected within the radius. Corresponding to the grid space, the position of neuron m is represented by the coordinates (lx m ,ly m ), the position of neuron n is represented by coordinates (lx n ,ly n ), then the distance between neuron m and neuron n is
[0056] From the neuron activity value formula, we can know that the grid in the forbidden state has its corresponding neuron activated by the external input signal I m =-E and the sum of the excitations transmitted by the connected neurons, then the corresponding neuron activity value is -1. The grid in the unsearched state has the corresponding neuron receiving the external input signal I m=E and the sum of the excitations transmitted by the connected neurons, the corresponding neuron activity value remains 1. The grid in the searched state has its corresponding neuron receiving the external input signal I m =0 and the sum of the excitations transmitted by the connected neurons, then the corresponding neuron activity value ranges from [0,1).
[0057] In another embodiment, the method for determining the comprehensive score of the neighborhood grid g in step 3 according to the activity value of the neuron corresponding to the neighborhood grid g at time t combined with the comprehensive discrete degree index is: according to AUV i Its own position information and any AUV j The position information of the AUV is determined i When moving to the neighborhood grid g, the AUV j The distance cost J d (t) and relative position And determine the AUV i and AUV j The dispersion index m d is the coefficient, m d ∈(-1,0). Among them, AUV i When moving to the neighborhood grid g, the AUV j The closer the distance, the greater the distance cost J d (t) is larger. AUV i With AUV j The greater the difference in direction angles, the greater the relative orientation The larger the integer parameter j∈[1,N] and j≠i.
[0058] According to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t), combined with other AUVs j The dispersion index Determine the comprehensive score F of the neighborhood grid g g (t) = z g (t)+f g (t), where the comprehensive dispersion index
[0059] During the multi-AUV collaborative search process, each AUV needs to consider its own state and the state of other AUVs when moving to the neighborhood grid to ensure that the movement paths between any two AUVs do not conflict and do not repeat. In one embodiment, please refer to Figure 4 The schematic diagram of multi-AUV coordinated motion is shown in Figure 2. i The AUV is determined by its own position information and the relative position relationship between the neighborhood grid g and the grid at the current time t. iThe heading angle of the movement to the neighborhood grid g According to AUV i Own location information and AUV j The position information of the AUV is determined i At the current time t, with AUV j The relative direction between According to the heading angle and relative direction Identify AUV i When moving to the neighborhood grid g, the AUV j Relative position of
[0060]
[0061] In one embodiment, according to the AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and the location coordinates of the neighborhood grid g (x g ,y g ), determine the AUV i The heading angle of the movement to the neighborhood grid g
[0062]
[0063] In one embodiment, according to the AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and AUV j The position coordinates of the grid at the current time t (l xj (t),l yj (t)), determine the AUV i With AUV j The relative direction between
[0064] The collaborative search of multiple AUVs should ensure that each AUV is responsible for a different area to avoid the problem of search area coverage. This application solves this problem by setting a distance penalty when the distance between AUVs is too close. i When moving to the neighborhood grid g, the AUV j The distance cost d ij (t) is AUV i When moving to the neighborhood grid g, the AUV j The distance between s It is a predetermined minimum distance threshold, and the minimum distance threshold can be customized. In this embodiment, the minimum distance threshold is set to a distance of 6 grids.
[0065] In order to verify the multi-AUV collaborative deep-sea cable search method proposed in this application, a simulation verification experiment was carried out on Matlab software. The seabed mission area of the AUV underwater working environment was set to a 100×100 discretized grid map, the parameters of the Glasius bionic neural network model were set to β=0.5, γ=0.3, the external input excitation E=30, m d =0.5, and the number of AUVs is set to 3. The multi-AUV collaborative deep-sea cable search method of the present application is used to search the seabed mission area, and the coverage curve of the seabed mission area search is obtained as follows: Figure 5 The coverage map of the seabed mission area is shown in Figure 6 As shown in the figure, the black area corresponds to the forbidden state, the white area corresponds to the unsearched state, and the gray area corresponds to the searched state. Figure 7 As shown, the neuron activity value in the black area is -1, corresponding to the grid in the forbidden state, the neuron activity value in the white area is 1, corresponding to the grid in the unsearched state, and the neuron activity value in the gray area is between 0 and 1, corresponding to the grid in the searched state. Figure 6 and Figure 7 From the comparison results, it can be seen that the neuron map and the coverage map are generally the same, which means that the state of the grid can be represented by the neuron activity value, and then the multi-AUV collaborative search method of the present application is used to select the movement path for each AUV at each moment, and finally the entire seabed mission area is searched, all grid maps are covered, and the submarine cable is found.
[0066] To further verify the effect of multi-AUV collaborative search in this application, a multi-AUV collaborative performance simulation verification experiment was conducted. The number of AUVs was set to 1, 3, and 5, and the coverage rate comparison curve of the seabed mission area search was as follows: Figure 8 When the number of AUVs is set to be larger, the search time is shorter, which verifies the effectiveness of multi-AUV collaborative search. It is possible to use multi-AUV collaborative search to reduce search time and improve the efficiency of submarine cable search operations.
[0067] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.
Claims
1. A multi-AUV collaborative deep-sea cable search method based on neural network, characterized in that: N AUVs establish communication connections between each other, and the integer parameter N≥2. The multi-AUV collaborative deep-sea cable search method includes: for any AUV i : The seabed mission area is rasterized to obtain a grid map, and a two-dimensional neural network model is generated according to the grid map using a Glasius bionic neural network, wherein neurons in the two-dimensional neural network model correspond one-to-one to grids of the grid map, and there is a connection relationship between neurons corresponding to two grids within a single-moment motion step of each AUV; At time t=0, the activity value of each neuron in the two-dimensional neural network model is initialized according to the grid state of each grid, and the position information of the current grid is initialized; the grid state of each grid is any one of a forbidden state, a searched state and a non-searched state; At any time t, the position information of the grids where other AUVs are located is obtained through the communication connection with other AUVs; for any neighboring grid g that is not searched within the single-time motion step of the AUV in the current grid, according to the AUV i Its own position information and the position information of other AUVs are used to evaluate the AUV when it moves to the neighborhood grid g. i The comprehensive dispersion index f between each other AUV g (t), and according to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t) Combined with the comprehensive dispersion index f g (t) Determine the comprehensive score F of the neighborhood grid g g (t), dispersion index f g (t) indicates that the higher the distribution dispersion between the AUVs and other AUVs, the higher the activity value of the neuron and the higher the comprehensive score; At time t+1, move to the neighborhood grid with the highest comprehensive score for detection, update the grid status of each grid, and update the activity value of each neuron at time t+1 according to the grid status of each grid and the connection relationship of each neuron until the search of the seabed mission area is completed.
2. The multi-AUV collaborative deep-sea cable search method according to claim 1 is characterized in that: Determining the comprehensive score of the neighborhood grid g includes: According to AUV i Its own position information and any AUV j The position information of the AUV is determined i When moving to the neighborhood grid g, the AUV j The distance cost J d (t) and relative position And determine the AUV i and AUV j The dispersion index between m d is the coefficient; where AUV i When moving to the neighborhood grid g, the AUV j The closer the distance, the greater the distance cost J d (t) The larger the AUV i With AUV j The greater the difference in direction angles, the greater the relative orientation The larger the value, the integer parameter j∈[1,N] and j≠i; According to the activity value z of the neuron corresponding to the neighborhood grid g at time t g (t), combined with other AUVs j The dispersion index between Determine the comprehensive score F of the neighborhood grid g g (t).
3. The multi-AUV collaborative deep-sea cable search method according to claim 2 is characterized in that: Identify AUV i When moving to the neighborhood grid g, the AUV j Relative position of include: According to AUV i The AUV is determined by its own position information and the relative position relationship between the neighborhood grid g and the grid at the current time t. i The heading angle of the movement to the neighborhood grid g According to AUV i Own location information and AUV j The position information of the AUV is determined i At the current time t, with AUV j The relative direction between According to the heading angle and relative direction Identify AUV i When moving to the neighborhood grid g, the AUV j Relative position of 4. The multi-AUV collaborative deep-sea cable search method according to claim 2 is characterized in that: AUV i When moving to the neighborhood grid g, the AUV j The distance cost d ij (t) is AUV i When moving to the neighborhood grid g, the AUV j The distance between s is the predetermined minimum distance threshold.
5. The multi-AUV collaborative deep-sea cable search method according to claim 3 is characterized in that: Determining the AUV i When moving to the neighborhood grid g, the AUV j Relative position of 6. The multi-AUV collaborative deep-sea cable search method according to claim 3 is characterized in that: Determining the AUV i The heading angle of the movement to the neighborhood grid g Includes: According to AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and the location coordinates of the neighborhood grid g (x g ,y g ), determine the AUV i The heading angle of the movement to the neighborhood grid g 7. The multi-AUV collaborative deep-sea cable search method according to claim 3 is characterized in that: Determining the AUV i At the current time t, with AUV j The relative direction between include: According to AUV i The position coordinates of the grid at the current time t (l xi (t),l yi (t)), and AUV j The position coordinates of the grid at the current time t (l xj (t),l yj (t)), determine the AUV i With AUV j The relative direction between 8. The multi-AUV collaborative deep-sea cable search method according to claim 2, characterized in that: Determine the comprehensive score F of the neighborhood grid g g (t) = z g (t)+f g (t), where the comprehensive dispersion index 9. The multi-AUV collaborative deep-sea cable search method according to claim 1, characterized in that: The updating of the activity value of each neuron at time t+1 according to the grid state of each grid and the connection relationship of each neuron includes for any neuron m: According to the grid state of the grid corresponding to neuron m, the external input signal I of neuron m is determined m ; When the grid state of the grid corresponding to neuron m is unsearched, I m =E; when the grid state of the grid corresponding to neuron m is forbidden, I m = -E; when the grid state of the grid corresponding to neuron m is the searched state, I m =0; E is a positive constant much greater than 1; According to the activity value z of any neuron n connected to neuron m at time t n (t), determine the activity value of neuron m at time t+1 Among them, [z n (t)] + =max(z n (t),0), N(m) is the set of neurons connected to neuron m, W mn is the connection weight between neuron m and neuron n and W mn =W nm ; Parameter β>
0.
10. The multi-AUV collaborative deep-sea cable search method according to claim 9, characterized in that: The connection weight W between neuron m and neuron n mn for: Where γ is a positive constant coefficient and γ∈(0,1], |mn| is the distance between neuron m and neuron n, and R is the range radius of neuron connection.
Citation Information
Cited By
Multi-AUV acousto-optic multimode collaborative search method based on deep reinforcement learning
CN119182460A
Under-actuated AUV (Autonomous Underwater Vehicle) trajectory planning method and system based on neurodynamic structure
CN119354204A
