Marine net cage unmanned ship intelligent inspection method based on buoy network and TDoA positioning

Abstract

The invention provides an ocean net cage unmanned ship intelligent inspection method based on a buoy network and TDoA positioning, and the method comprises the steps: constructing an intelligent buoy network in an ocean net cage culture area, and obtaining a refined grid map; performing global path planning by adopting an improved A * algorithm to generate a global reference path; the unmanned ship transmits a UWB signal to the navigation buoy, the current position of the unmanned ship is solved based on a TDoA principle and an LM algorithm, and a TDoA positioning result is obtained; the self-adaptive UKF is used for fusing the state information of the unmanned ship and the TDoA positioning result, and high-precision position estimation of the unmanned ship is obtained; and finally, the unmanned ship executes an autonomous inspection task according to the global reference path, and precise path tracking and calibration are realized by utilizing PID (Proportion Integration Differentiation) control. According to the invention, autonomous, safe and efficient inspection of the unmanned ship in a net cage dense area can be realized, the intelligent level of breeding management is improved, and the system is especially suitable for application scenes with high positioning precision requirements.

Description

Technical Field

[0001] The present invention relates to the technical fields of unmanned ship path planning and marine aquaculture management, and more specifically, to an intelligent inspection method for a marine cage unmanned ship based on a buoy network and TDoA (Time Difference of Arrival) positioning. Background Art

[0002] In the management of marine cage aquaculture, regular inspections are crucial for ensuring aquaculture safety, promptly detecting and handling potential problems. Traditional inspection methods mainly rely on manually driving ships, which are inefficient, costly, and greatly affected by human factors and weather conditions.

[0003] Although there have emerged some technologies for inspecting using unmanned ships, the existing unmanned ship inspection technologies still face many challenges in the cage-intensive area, making it difficult for traditional inspection methods to meet the needs of autonomous inspection in the cage aquaculture area. Specifically, the existing unmanned ship inspection technologies have the following problems:

[0004] 1) Limitations of Visual SLAM (Simultaneous Localization and Mapping): Visual SLAM highly depends on visual features. In cases of light changes, water surface reflection, lack of features, etc., its performance will significantly decline, making it difficult to ensure positioning accuracy and robustness.

[0005] 2) Limitations of lidar: Lidar is vulnerable to water surface reflection, scattering, etc., resulting in a decline in the quality of point cloud data, affecting positioning and mapping accuracy, and having a high cost.

[0006] 3) Limitations of GPS: In the cage-intensive area, GPS signals are vulnerable to occlusion and reflection, resulting in a decline in positioning accuracy or even signal loss, and unable to meet the centimeter-level positioning requirements.

[0007] 4) Limitations of traditional path planning: Rule-based path planning methods rely on accurate modeling and static assumptions, making it difficult to cope with dynamic changes in cage states or emergencies, lacking flexibility and adaptability.

[0008] 5) Limitations of existing buoy systems: Existing buoy systems have a single function, lack the ability to interact with unmanned ships, cannot provide sufficient information support, and have deficiencies in data transmission, energy consumption, and intelligence.

[0009] Therefore, it is necessary to develop an unmanned ship autonomous inspection scheme for the cage aquaculture area that can overcome the above limitations and achieve high precision, high robustness, safety, and reliability. Summary of the Invention

[0010] To overcome the defects of low positioning accuracy, susceptibility to environmental interference, unsafe and irregular path planning existing in the above-mentioned existing unmanned ship inspection technologies, the present invention provides an intelligent inspection method for ocean cage unmanned ships based on a buoy network and TDoA positioning, which can realize autonomous, safe and efficient inspection of unmanned ships in the cage-intensive area, improve the intelligent level of aquaculture management, and is especially suitable for application scenarios with high positioning accuracy requirements.

[0011] To solve the above technical problems, the technical solution of the present invention is as follows:

[0012] An intelligent inspection method for ocean cage unmanned ships based on a buoy network and TDoA positioning, comprising the following steps:

[0013] S1: Construct an intelligent buoy network in the ocean cage aquaculture area, including a number of navigation buoys and a number of boundary buoys; the navigation buoys are used to provide UWB TDoA positioning reference and virtual navigation channels for the unmanned ship; the boundary buoys are used to identify the boundary of the aquaculture area;

[0014] S2: Based on the intelligent buoy network, construct a refined grid map of the aquaculture area, and divide the aquaculture area into passable areas, cage no-go areas, buoy no-go areas and areas outside the boundary;

[0015] S3: Based on the refined environment map, use an improved A* algorithm for global path planning to generate a global reference path; the cost function of the improved A* algorithm introduces additional costs based on grid types and buoy-related costs;

[0016] S4: The unmanned ship emits UWB signals, each of the navigation buoys receives the UWB signals and records the timestamps, calculates the distance differences from the unmanned ship to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt algorithm to solve the current position of the unmanned ship to obtain the TDoA positioning result;

[0017] S5: The unmanned ship uses the sensors carried by itself to collect its own state information in real time, and fuses its own state information and the TDoA positioning result through an adaptive UKF to obtain a high-precision position estimate of the unmanned ship;

[0018] S6: The unmanned ship executes an autonomous inspection task according to the global reference path and sends the inspection data to the control center on the shore in real time. During the inspection process, based on the deviation between the high-precision position estimate of the unmanned ship and the global reference path, the heading of the unmanned ship is calibrated in real time based on a PID controller to achieve accurate path tracking and calibration.

[0019] Preferably, in the step S1, the navigation buoy is equipped with a UWB transceiver module, a GPS positioning module, a LoRa module, and an environmental monitoring sensor, and uses the IEEE 1588 precision time protocol for clock synchronization to provide the UWB TDoA positioning reference, virtual navigation channel, and environmental monitoring data for the unmanned boat;

[0020] The boundary buoy is equipped with a UWB transceiver module, a GPS positioning module, a LoRa module, and an environmental monitoring sensor, and is used to identify the boundary of the aquaculture area and provide environmental monitoring data;

[0021] The unmanned boat is equipped with a UWB transceiver module, a GPS positioning module, an IMU sensor, a controller, an adaptive UKF multi-sensor fusion module, a PID path tracking module, a map construction module, and a communication module, and is used to execute autonomous inspection tasks and achieve information interaction, high-precision positioning, and precise path tracking and calibration.

[0022] Preferably, the navigation buoy also obtains a rough position using the GPS positioning module, and performs UWB TDoA measurements with surrounding navigation buoys and boundary buoys at known positions, and uses the least squares method and the RANSAC algorithm for cooperative positioning to continuously optimize its position estimation.

[0023] Preferably, in the step S2, the divided areas of the refined grid map include:

[0024] Occupied cage no-go area: including all grids within a preset range around the occupied cages;

[0025] Empty cage no-go area: including all grids within a preset range around the empty cages;

[0026] Buoy no-go area: including all grids within a certain range around the navigation buoy or the boundary buoy;

[0027] Area outside the boundary: including all grids outside the effective range of the aquaculture area defined by the boundary buoy;

[0028] Passable area: including all other grids except the above no-go areas and the area outside the boundary.

[0029] Preferably, in the step S3, the cost function of the improved A* algorithm includes:

[0030] f(n) = g(n) + h(n) + c(n) + b(n)

[0031] b(n) = b reward (n) + b penalty boundary (n)

[0032] ​Among them, f(n) is the total cost of moving from the starting point to the current nth grid node; g(n) is the actual movement cost of moving from the starting point to the current nth grid node; h(n) is the heuristic estimated cost of moving from the current nth grid node to the target grid node; c(n) is the additional cost based on the grid type. For the grids in the no-go area and outside the boundary, c(n)=∞, and for the passable area, c(n)=0; b(n) is the buoy-related cost, b reward (n) is the reward for approaching the navigation buoy, b penalty (n) is the penalty for deviating from the center line of the virtual navigation channel; b boundary (n) is the penalty for moving away from the boundary buoy;

[0033] The said b reward (n) is expressed as:

[0034] b reward (n)=-k·max(0, R - d)

[0035] Among them, d is the distance from the current nth grid node to the nearest navigation buoy; R is the influence range radius of the navigation buoy; k is the reward coefficient;

[0036] The said b penalty (n) is expressed as:

[0037]

[0038] Among them, d channel is the vertical distance from the current nth grid node to the nearest center line of the virtual navigation channel; d1, d2, and d3 are the first, second, and third distance thresholds respectively, used to divide different deviation degrees; α1, α2, and α3 are the first, second, and third penalty coefficients respectively, and α1 < α2 < α3, used to control the penalty intensity of different deviation degrees to ensure that the farther the deviation, the greater the cost;

[0039] The said b boundary (n) is expressed as:

[0040] b boundary (n)=β·max(0, D - d bound )

[0041] Among them, d bound is the distance from the current nth grid node to the nearest boundary buoy, β is the penalty coefficient, and D is the boundary buoy safety distance threshold.

[0042] Preferably, in the step S4, the unmanned ship emits a UWB signal, and N said navigation buoys receive the UWB signal and record the timestamp t rx,i , where N is the total number of navigation buoys;

[0043] Preprocess the originally acquired timestamp t rx,i including time synchronization, antenna delay calibration, and jitter suppression to obtain a high-precision timestamp t'. rx,i ;

[0044] Select the j-th navigation buoy as the reference buoy, and calculate the signal arrival time difference of other navigation buoys relative to the reference buoy, expressed as:

[0045] Δt j,i = t′ rx,i - t′ rx,j

[0046] where Δt j,i is the signal arrival time difference of the i-th navigation buoy relative to the reference buoy;

[0047] Calculate the distance difference from the unmanned ship to each navigation buoy according to the following formula:

[0048] Δd j,i = c eff ·Δt j,i

[0049] where Δd j,i is the distance difference from the unmanned ship to the i-th navigation buoy and to the reference buoy; c eff is the effective propagation speed of the signal in the propagation medium;

[0050] Construct at least two distance difference equations according to the following formula to construct a nonlinear equation system:

[0051]

[0052] where (x, y) is the current position of the unmanned ship; (X i , Y i ) is the position of the i-th navigation buoy; (X j , Y j ) is the position of the reference buoy;

[0053] Use the Levenberg-Marquardt algorithm to iteratively solve the nonlinear equation system to obtain the current position (x, y) of the unmanned ship and obtain the TDoA positioning result.

[0054] Preferably, in the step S5, the self-state information of the unmanned ship collected in real time by the self-carried positioning sensor at least includes: GPS positioning data and IMU sensor data;

[0055] Fuse the self-state information and the TDoA positioning result through an adaptive UKF, including the prediction step and the update step executed in sequence;

[0056] In the prediction step, the state estimate of the unmanned ship at the next moment is predicted based on the adaptive UKF algorithm according to the IMU sensor data at the current moment; in the update step, the residual of the TDoA positioning result and the residual of the GPS positioning data are calculated respectively, and the weighted square sum of the residuals of the two is further calculated. The adaptive weight is finally fused and updated, and the updated state estimate is used as the high-precision position estimate of the unmanned ship.

[0057] Preferably, in step S6, according to the deviation between the high-precision position estimate of the unmanned ship and the global reference path, the unmanned ship heading is calibrated in real time based on a PID controller, including the following steps:

[0058] Search the global reference path for the distance from the current position of the unmanned ship to p k The nearest point p ref ; Calculate the current position p of the unmanned ship k To point p ref The vertical distance of the tangent line of the path is taken as the lateral deviation e d ; Calculate the current unmanned ship heading ψ k With point p ref The tangent direction of the path ψ ref The angle difference is taken as the heading deviation e ψ ;

[0059] The lateral deviation e d and heading deviation e ψ Common input PID controller, the output of the PID controller is the control instruction, including at least the desired rudder angle δ cmd ; The control law of the PID controller is expressed as:

[0060] δ cmd =delta d +delta ψ

[0061]

[0062] Among them, delta d Indicates the lateral deviation control amount; delta ψ Indicates the heading deviation control amount; K p,d , K i,d and K d,d Respectively represent the proportional gain, integral gain and differential gain of lateral deviation control; K p,ψ and K d,ψ Represents the proportional gain and derivative gain of heading deviation control.

[0063] Preferably, the step S6 further comprises: adaptively adjusting a gain parameter of the PID controller based on the positioning uncertainty or adaptive weight of the adaptive UKF;

[0064] The gain parameters of the PID controller are adaptively adjusted based on the positioning uncertainty of the adaptive UKF, including: calculating the positioning uncertainty index using the covariance matrix provided by the adaptive UKF algorithm; when the positioning uncertainty index is greater than the preset uncertainty threshold, reducing the gain parameters of the PID controller; otherwise, increasing the gain parameters of the PID controller.

[0065] The gain parameters of the PID controller are adaptively adjusted based on the adaptive weights, including: obtaining the adaptive weights of the TDoA positioning result and the GPS positioning data; when any one of the adaptive weights is less than the preset weight threshold, reducing the gain parameters of the PID controller; otherwise, increasing the gain parameters of the PID controller.

[0066] Preferably, the method further includes: when an abnormal situation occurs during the inspection, the unmanned vessel triggers actions such as decelerating and hovering, returning, or dynamic obstacle avoidance, and sends an alarm to the control center, waiting for manual intervention.

[0067] The abnormal situation includes at least any one of system failure, communication interruption, unreliable positioning result, detected obstacle, approaching the boundary, and low battery.

[0068] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0069] The present invention provides an intelligent inspection method for an unmanned vessel of a marine cage based on a buoy network and TDoA positioning. First, an intelligent buoy network is constructed in the marine cage aquaculture area, and then a refined grid map of the aquaculture area is constructed based on the intelligent buoy network, dividing the aquaculture area into a passable area, a cage restricted area, a buoy restricted area, and an area outside the boundary; based on the refined environmental map, an improved A* algorithm is used for global path planning to generate a global reference path; the unmanned vessel emits UWB signals, and each navigation buoy receives the UWB signals and records the timestamps, calculates the distance differences from the unmanned vessel to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt algorithm to solve the current position of the unmanned vessel to obtain the TDoA positioning result; then the unmanned vessel uses the sensors carried by itself to collect its own state information in real time, and fuses its own state information and the TDoA positioning result through adaptive UKF to obtain a high-precision position estimate of the unmanned vessel; finally, the unmanned vessel executes the autonomous inspection task according to the global reference path and sends the inspection data to the control center on the shore in real time. During the inspection process, based on the deviation between the high-precision position estimate of the unmanned vessel and the global reference path, the heading of the unmanned vessel is calibrated in real time based on the PID controller to achieve accurate path tracking and calibration.

[0070] The present invention has the following beneficial effects:

[0071] 1) High-precision positioning: By adopting the UWB TDoA positioning technology and combining it with the collaborative positioning of navigation buoys and the Levenberg-Marquardt algorithm, centimeter-level positioning accuracy can be achieved, which is significantly better than GPS positioning alone and is not easily affected by water surface reflection and occlusion, capable of meeting the refined inspection requirements;

[0072] 2) High-robustness positioning: By adaptively fusing TDoA and various sensor data with UKF, it can effectively suppress sensor noise and environmental interference, improve the robustness and reliability of the positioning system, and ensure positioning accuracy even when the GPS signal is poor;

[0073] 3) Safe and reliable path planning: Based on the intelligent buoy network, a refined environmental map is constructed, and the improved A* algorithm is used for global path planning, forcibly avoiding obstacles such as cages and buoys, and restricting within the boundary of the aquaculture area, effectively reducing the collision risk and ensuring the safe navigation of the unmanned ship;

[0074] 4) Path regularity and predictability: Using the virtual channels defined by navigation buoys to guide the path, making the inspection path more regular, improving the predictability of the path, and facilitating remote monitoring and management;

[0075] 5) Autonomous and efficient inspection: The unmanned ship can autonomously execute inspection tasks without manual intervention, reducing labor costs and improving inspection efficiency and coverage;

[0076] 6) Multifunctional integrated intelligent buoy network: The intelligent buoy network integrates multiple functions such as UWB positioning reference, virtual navigation channels, environmental monitoring, and boundary warning, providing comprehensive information support for the unmanned ship;

[0077] 7) Low-power design: The unmanned ship only needs to emit UWB signals once, which are received by multiple buoys simultaneously, significantly reducing the power consumption at the unmanned ship end and extending the battery life. Description of the Drawings

[0078] Figure 1 It is a flowchart of an intelligent inspection method for an ocean cage unmanned ship based on a buoy network and TDoA positioning provided in Embodiment 1.

[0079] Figure 2 It is a hardware architecture diagram of an intelligent inspection method for an ocean cage unmanned ship based on a buoy network and TDoA positioning provided in Embodiment 2.

[0080] Figure 3 It is an example diagram of the layout of the intelligent buoy network in the cage aquaculture area provided in Embodiment 2.

[0081] Figure 4 It is a schematic diagram of the collaborative positioning of navigation buoys provided in Embodiment 2.

[0082] Figure 5 Schematic diagram of the path planning result of the improved A* algorithm provided in Embodiment 2.

[0083] Figure 6 Flow chart of multi-sensor fusion based on adaptive UKF provided in Embodiment 2.

[0084] Figure 7 Schematic diagram of the path tracking control of the unmanned boat based on PID provided in Embodiment 2. Specific implementation manners

[0085] The accompanying drawings are only for illustrative purposes and should not be construed as a limitation to this application.

[0086] To better illustrate this embodiment, some components in the accompanying drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product.

[0087] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0088] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0089] Embodiment 1

[0090] As Figure 1 shown, this embodiment provides an intelligent inspection method for an unmanned boat of an ocean cage based on a buoy network and TDoA positioning, including the following steps:

[0091] S1: Construct an intelligent buoy network in the ocean cage aquaculture area, including a number of navigation buoys and a number of boundary buoys; the navigation buoys are used to provide UWB TDoA positioning references and virtual navigation channels for the unmanned boat; the boundary buoys are used to identify the boundaries of the aquaculture area.

[0092] S2: Construct a refined grid map of the aquaculture area based on the intelligent buoy network, and divide the aquaculture area into a passable area, a cage restricted area, a buoy restricted area, and an area outside the boundary.

[0093] S3: Based on the refined environmental map, use the improved A* algorithm for global path planning to generate a global reference path; the cost function of the improved A* algorithm introduces additional costs based on grid types and buoy-related costs.

[0094] S4: The unmanned boat emits UWB signals, each of the navigation buoys receives the UWB signals and records timestamps, calculates the distance differences from the unmanned boat to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt algorithm to solve the current position of the unmanned boat to obtain the TDoA positioning result.

[0095] S5: The unmanned ship uses the sensors it carries to collect its own status information in real time, and fuses its own status information and the TDoA positioning result through an adaptive UKF to obtain a high-precision position estimate of the unmanned ship.

[0096] S6: The unmanned ship performs autonomous inspection tasks according to the global reference path and sends the inspection data to the control center on the shore in real time. During the inspection process, based on the deviation between the high-precision position estimate of the unmanned ship and the global reference path, the heading of the unmanned ship is calibrated in real time based on a PID controller to achieve precise path tracking and calibration.

[0097] In the specific implementation process, first, an intelligent buoy network is constructed in the marine cage aquaculture area. The intelligent buoy network consists of navigation buoys and boundary buoys. The navigation buoys use the IEEE 1588 Precision Time Protocol (PTP) for wireless synchronization to provide UWB TDoA positioning benchmarks and virtual navigation channels; the boundary buoys are used to identify the aquaculture area boundary and provide boundary warnings; at the same time, the navigation buoys perform UWB TDoA measurements with the surrounding buoys with known positions and use the least squares method for cooperative positioning to optimize their own position estimates.

[0098] Then, based on the layout information of the intelligent buoy network (buoy positions, precise positions of cages, dimensions and status, boundary ranges), a refined grid map of the aquaculture area is constructed, and the aquaculture area is divided into passable areas, cage no-go areas, buoy no-go areas, and areas outside the boundary to provide a basis for path planning and obstacle avoidance; the cage no-go areas and buoy no-go areas are defined by adding a safety distance based on the center positions of the cages / buoys.

[0099] Based on the refined environment map, an improved A* algorithm is used for global path planning to generate a global reference path, so as to ensure that the unmanned ship sails on a safe and regular path; the improved A* algorithm enables the unmanned ship to forcibly avoid all no-go areas defined by cages and buoys and restricts it to sail within the effective area defined by the boundary buoys; at the same time, the center line of the virtual channel defined by the navigation buoys is used to guide the path.

[0100] After that, the unmanned ship periodically or on demand emits UWB (Ultra-Wideband) signals. Each navigation buoy receives the UWB signal and records the timestamp, calculates the distance difference from the unmanned ship to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt (LM) algorithm to solve the current position of the unmanned ship to obtain the TDoA positioning result.

[0101] After that, the unmanned ship uses its own sensors to collect its own status information in real time, and fuses its own status information and the TDoA positioning result through an adaptive UKF to obtain a high-precision position estimate of the unmanned ship, thereby improving the positioning accuracy and robustness of the unmanned ship and providing a positioning quality assessment; the state vector of the adaptive UKF algorithm includes position, speed, attitude quaternion, and accelerometer bias, and is iteratively calculated through the prediction step and the update step to achieve the optimal estimation of the unmanned ship's state;

[0102] Finally, the unmanned ship executes the autonomous inspection task according to the global reference path and sends the inspection data to the control center on the shore in real time;

[0103] During the inspection process, based on the deviation between the high-precision position estimate of the unmanned ship (from the adaptive UKF fusion positioning result) and the global reference path (including lateral deviation and heading deviation), the heading of the unmanned ship is calibrated in real time based on the PID controller to achieve precise path tracking and calibration; when the unmanned ship approaches the navigation buoy, the observation noise of the TDoA measurement value in the adaptive UKF is dynamically adjusted to increase the weight of the TDoA positioning result for calibration, further improving the positioning accuracy;

[0104] The unmanned ship sails autonomously according to the above steps, executes the cage inspection task, and at the same time interacts with the intelligent buoy network to obtain environmental data and boundary warning information and sends them to the control center in real time, so as to achieve safe and efficient autonomous inspection;

[0105] This method can realize the autonomous, safe and efficient inspection of unmanned ships in the intensive cage area, improve the intelligent level of aquaculture management, and is especially suitable for application scenarios with high positioning accuracy requirements.

[0106] Embodiment 2

[0107] Based on Embodiment 1, this embodiment provides an intelligent inspection method for an ocean cage unmanned ship based on a buoy network and TDoA positioning.

[0108] In the specific implementation process, as Figure 2 shown, first introduce the hardware architecture used in this embodiment:

[0109] 1) The intelligent buoy network consists of navigation buoys and boundary buoys arranged according to rules. The navigation buoys are used to provide UWB TDoA positioning benchmarks, virtual navigation channels, environmental data, and their own precise positions; the boundary buoys are used to identify the boundaries of the aquaculture area, provide boundary warnings, and can assist in positioning;

[0110] 2) The unmanned ship is equipped with a UWB transceiver module, a GPS positioning module, an IMU sensor, a controller, a multi-sensor fusion module, a path tracking module, a map building module, and a communication module; the UWB transceiver module is used to transmit TDoA positioning signals; the GPS positioning module and the IMU sensor are used to provide auxiliary positioning and attitude information; the controller is used to execute path planning, path tracking, and task management; the multi-sensor fusion module is used to fuse the TDoA positioning results, GPS data, and IMU data; the path tracking module is used to control the movement of the unmanned ship according to the real-time position and the reference path; the communication module is used to interact with the intelligent buoy network and an optional shore-based control system for information.

[0111] 3) The control center is used for task configuration, remote monitoring, data storage, and analysis.

[0112] In this embodiment, the intelligent buoy network consists of N intelligent buoys (N is determined according to the size, shape, cage layout, and accuracy requirements of the actual aquaculture area), and is arranged in the cage aquaculture area according to specific rules; the buoys are divided into two types: navigation buoys (as shown by the circular icon in Figure 3 ) and boundary buoys (as shown by the triangular icon in Figure 3 ); the buoys adopt a streamlined design to reduce wind resistance and water resistance; and use counterweights to lower the center of gravity and improve stability; the outer shell is made of corrosion-resistant materials (e.g., HDPE, fiberglass), and metal components are made of stainless steel or undergo anti-corrosion treatment; the outer shell is coated with an anti-fouling coating to reduce the attachment of marine organisms; a suitable anchor chain + sinker (or suction anchor, etc.) method is used to reliably fix the buoy at a predetermined position, considering water depth, bottom sediment, and environmental forces.

[0113] The arrangement method of the buoys adopts an arrangement method optimized considering the Geometric Dilution of Precision (GDOP), aiming to ensure good positioning accuracy in most passable areas within the aquaculture area; such an arrangement method can better cover the rectangular cage aquaculture area, ensuring that the unmanned ship can receive a sufficient number (usually ≥ 3) of navigation buoy signals in most areas; in the case of partial buoy failure, it can still ensure a certain positioning accuracy; compared with a denser arrangement method, while ensuring accuracy, it reduces the deployment cost; at the same time, it has a clear navigation structure, clearly defining the boundary and recommended navigation channels, facilitating path planning and guidance; in addition, the relative geometric configuration of the buoys directly affects the accuracy of TDoA positioning, and this influence is usually quantified by the GDOP value. The smaller the GDOP value, the higher the positioning accuracy; the design goal of the buoy arrangement method adopted in this embodiment is to keep the GDOP value in most passable areas within the aquaculture area at a low level (e.g., expected to be less than a certain threshold, such as 5 or 3), so as to meet the requirements of centimeter-level positioning accuracy.

[0114] The calculation of GDOP is based on the geometric relationship between the buoy positions and the positions of the unmanned vessels. The GDOP value reflects the magnification factor of the position error, which can quantify the impact of the buoy geometric configuration on the positioning accuracy. By optimizing the buoy layout reasonably (such as adjusting the number, position, and spacing of the buoys), the control of GDOP can be achieved to reach the expected positioning accuracy. In the edge or corner areas of the aquaculture zone, the GDOP value may increase naturally, and the accuracy of these areas can be compensated by increasing the number of boundary buoys, adjusting the buoy positions, or relying more on other positioning means for fusion (such as GPS).

[0115] As Figure 3 shown, an exemplary arrangement is for a rectangular aquaculture area with a two - row and four - column circular cage layout, adopting a three - row and five - column buoy arrangement; in Figure 3 it, the cage layout is as follows:

[0116] The origin of the coordinate system (0, 0) is located at the lower - left corner of the aquaculture area, the x - axis is to the right (east - west direction), and the y - axis is upward (north - south direction); the cage diameter is 30m, arranged in two rows and four columns, with a quantity of 8, the row spacing is 60m, and the column spacing is 60m; the center coordinates of the first cage (lower - left corner): (45, 30); the list of the center coordinates of all cages:

[0117] C1(45, 30), C2(105, 30), C3(165, 30), C4(225, 30) (the first row);

[0118] C5(45, 90), C6(105, 90), C7(165, 90), C8(225, 90) (the second row);

[0119] The buoy layout is as follows:

[0120] The buoys in the first row and the third row (boundary buoys): are located on the center line of the outer channel of the cages or on the boundary line of the aquaculture area, maintaining a safe distance from the edge of the nearest cage (for example, at least 15 meters);

[0121] The buoys in the second row (navigation buoys): are located on the center line between the two rows of cages, maintaining a safe distance from the edges of the cages on both sides (for example, 15 meters each); the navigation buoys in this row define a main virtual channel center line;

[0122] The buoy spacing: within the row and between the columns can be determined according to the coverage requirements and the principle of optimizing the TDoA geometric accuracy (GDOP); for example, in the Figure 3 shown example scenario (a 240m×120m aquaculture area), the row spacing can be set to 60m, and the column spacing can be set to 60m; the buoy coordinates can be defined as follows:

[0123] B1(0, 120), B2(60, 120), B3(120, 120), B4(180, 120), B5(240, 120) (Boundary);

[0124] B6(0, 60), B7(60, 60), B8(120, 60), B9(180, 60), B10(240, 60) (Navigation);

[0125] B11(0, 0), B12(60, 0), B13(120, 0), B14(180, 0), B15(240, 0) (Boundary);

[0126] In this embodiment, each buoy is equipped with a microcontroller, a LoRa module, a UWB module, a positioning module (the navigation buoy uses GPS / UWB integrated positioning, where GPS provides the initial rough position and time reference, and UWB is used for cooperative positioning with other buoys to improve its own position accuracy and serves as a TDoA positioning base station, and the boundary buoy uses a GPS module for positioning), sensors (including environmental sensors such as temperature, pH value, and dissolved oxygen sensors; an ultrasonic sensor can also be configured to set a detection distance to sense the approach of the unmanned boat, which is used to trigger specific interactions and assist in short - range obstacle avoidance when the unmanned boat approaches), LED indicators (used for status indication and visual recognition under night or low visibility conditions), and a power module (a combination of a solar panel and a lithium iron phosphate battery can be used); LoRaWAN protocol or other suitable low - power wide - area network protocols can be used between buoys or between a buoy and an unmanned boat;

[0127] In this embodiment, it is also necessary to synchronize the communication of the intelligent buoy network, such as Figure 4 As shown, data exchange (such as its own precise position information, environmental monitoring data, status information, timestamp, etc.) occurs between buoys through LoRa; IEEE 1588 Precision Time Protocol (PTP) wireless synchronization is adopted between navigation buoys to achieve high - precision clock synchronization. By exchanging PTP synchronization messages and using mechanisms such as hardware timestamps, two - way timestamp exchange, and path delay compensation, a synchronization accuracy better than 5 nanoseconds is achieved, which is the basis for TDoA positioning;

[0128] Specifically, the navigation buoy can obtain a rough position using GPS and perform UWB TDoA measurements with surrounding buoys with known positions (including boundary buoys and other accurately positioned navigation buoys), and use the least - squares method for cooperative positioning to continuously optimize its own position estimate. The steps are as follows:

[0129] a) First, based on constructing an error equation: By measuring the distance differences between each navigation buoy and the buoy with known precise position using UWB, an error equation based on the distance differences is constructed;

[0130] b) When there are enough buoy markers with known positions around the navigation buoy, there will be multiple error equations. Combine all the error equations into a vector F;

[0131] c) Linearize the error equations: ΔF = J·ΔX, where ΔX is the position increment and J is the Jacobian matrix; Solve for ΔX = (J T ·J) -1 ·J T ·(-F) through the least squares method, and iteratively update the position X of the navigation buoy as X = X GPS +ΔX, where X GPS is the GPS position until convergence; Meanwhile, adopt the RANSAC (Random Sample Consensus) algorithm to detect and eliminate abnormal measurement values (gross errors) caused by multipath, etc., to improve the robustness of cooperative positioning;

[0132] For example, assume that the initial position estimate of the navigation buoy B7(60, 60) has a deviation, which is (61, 62, 0); Through UWB ranging with the surrounding buoy markers with known accurate positions (such as B1(0, 120), B2(60, 120), B6(0, 60), B8(120, 60), B12(60, 0)), a series of distance measurement values are obtained; Use these measurement values and the accurate coordinates of the known buoy markers to construct an error equation system; Through the least squares method and the RANSAC algorithm (eliminating gross errors that may be caused by multipath), iteratively solve for the position increment ΔX, and continuously update the position estimate of B7 until it converges to a value close to its true coordinates (60, 60, 0);

[0133] Then, based on the layout information of the intelligent buoy network (buoy positions, accurate positions of the cages, dimensions and states, boundary ranges), construct a refined grid map of the aquaculture area, divide the aquaculture area into passable areas, cage restricted areas, buoy restricted areas, and areas outside the boundary, providing a basis for path planning and obstacle avoidance; The cage restricted areas and buoy restricted areas are defined by adding a safety distance based on the center positions of the cages / buoys, specifically as follows:

[0134] Occupied cage restricted area: The grid within a certain range (cage radius + preset safety distance, the safety distance is greater than or equal to 5 meters) centered on the occupied cage;

[0135] Vacant cage restricted area: The grid within a certain range (cage radius + preset safety distance, the safety distance is greater than or equal to 5 meters) centered on the vacant cage;

[0136] Buoy restricted area: The grid within a certain range (buoy radius + preset safety distance, the safety distance is greater than or equal to 2 meters) centered on any intelligent buoy (navigation buoy or boundary buoy); This area is used to prevent collisions with the buoy;

[0137] Off - boundary area: grids located outside the effective range of the aquaculture area defined by boundary buoys;

[0138] Passable area: all other grids that do not belong to any of the above no - go areas or off - boundary areas;

[0139] For example, in Figure 3 the shown scenario, assume that a path needs to be planned from the docking point position A1(60, 30) near cage 1 to the docking point position A2(225, 90) near cage 8; the diameter of the cage is 30m (radius 15m), and the radius of the buoy is assumed to be 0.5m; example of the construction method: set the safety distance of the cage to 5m and the safety distance of the buoy to 2m;

[0140] Cage no - go area: a circular area with a radius of 15m + 5m = 20m centered on the center of each cage; for example, the grids within a radius of 20m centered on the cage (45, 30) are marked as no - go;

[0141] Buoy no - go area: a circular area with a radius of 0.5m + 2m = 2.5m centered on the center of each buoy; for example, the grids within a radius of 2.5m centered on the navigation buoy B7(60, 60) are marked as no - go;

[0142] Off - boundary area: the area where x < 0, x > 240, y < 0, y > 120;

[0143] After that, based on the refined environmental map, an improved A* algorithm is used for global path planning to generate a global reference path, so as to ensure that the unmanned ship sails on a safe and regular path; the improved A* algorithm enables the unmanned ship to forcibly avoid all no - go areas defined by cages and buoys and restricts it to sail within the effective area defined by boundary buoys; at the same time, the virtual channel center line defined by navigation buoys is used to guide the path;

[0144] In this embodiment, the cost function of the improved A* algorithm includes:

[0145] f(n)=g(n)+h(n)+c(n)+b(n)

[0146] b(n)=b reward (n)+b penalty (n)+b boundary (n)

[0147] Among them, f(n) is the total cost of moving from the starting point to the current nth grid node; g(n) is the actual movement cost of moving from the starting point to the current nth grid node; h(n) is the heuristic estimated cost of moving from the current nth grid node to the target grid node; c(n) is the additional cost based on the grid type. For grids in no-go areas and outside the boundary, c(n) = ∞ (indicating that it is absolutely impassable), and for passable areas, c(n) = 0; b(n) is the buoy-related cost, b reward (n) is the reward for approaching the navigation buoy, b penalty (n) is the penalty for deviating from the center line of the virtual navigation channel; b boundary (n) is the penalty for moving away from the boundary buoy;

[0148] The reward mechanism can make the path more inclined to pass through areas with better positioning signals, which helps to improve the navigation accuracy, b reward (n) is expressed as:

[0149] b reward (n) = -k·max(0, R - d)

[0150] Among them, d is the distance from the current nth grid node to the nearest navigation buoy; R is the influence range radius of the navigation buoy, for example, set to 15m, which is determined according to the positioning accuracy of the navigation buoy and the control accuracy of the unmanned ship; k is the reward coefficient, for example, set to 0.1, which controls the reward intensity for approaching the navigation buoy. The larger the k value, the more the path tends to approach the navigation buoy;

[0151] In order to guide the unmanned ship to preferentially travel along the "virtual navigation channel" defined by the navigation buoys, a penalty cost is added to the grids that deviate from these channels; in this embodiment, the virtual navigation channel is an area defined by a series of channel center lines (usually line segments connecting adjacent navigation buoys) and the channel width. For example, in Figure 3 it, the main virtual channel center line is the line segment connecting B6 - B7 - B8 - B9 - B10, that is, y = 60, 0 ≤ x ≤ 240; if the channel width is set to 4m, the virtual navigation channel area is within the area 2m on both sides of the y = 60 center line, that is, the area where 58 ≤ y ≤ 62;

[0152] This embodiment adds a penalty cost b penalty (n) to the nodes that deviate from the virtual channel area, which is expressed as:

[0153]

[0154] Among them, d channelis the perpendicular distance from the current nth grid node to the center line of the nearest virtual navigation channel; d1, d2, and d3 are the first, second, and third distance thresholds respectively, used to divide different degrees of deviation; α1, α2, and α3 are the first, second, and third penalty coefficients respectively, and α1 < α2 < α3, used to control the penalty intensity of different degrees of deviation to ensure that the greater the deviation, the greater the cost;

[0155] This embodiment also introduces a penalty for being far from the boundary buoy. If the distance from node n to any boundary buoy is less than the safety distance threshold D (for example, 5m), the penalty b is increased boundary (n) is expressed as:

[0156] b boundary (n) = β · max(0, D - d bound )

[0157] where d bound is the distance from the current nth grid node to the nearest boundary buoy, and β is the penalty coefficient;

[0158] As Figure 5 shown, for the path planning from the docking point position A1(60, 30) near cage 1 to the docking point position A2(225, 90) near cage 8, the improved A* algorithm is applied as follows:

[0159] 1) Construct the environmental map: Divide the 240m x 120m aquaculture area into grids of, for example, 5m x 5m; Mark the prohibited grids according to the positions of the cages (radius 15m + safety distance 5m) and buoys (radius 0.5m + safety distance 2m);

[0160] 2) Initialize: Create an open list and a closed list; The starting point A1(60, 30) corresponds to the grid (12, 6); Add it to the open list; g(A1) = 0;

[0161] h(A1) = Manhattan distance ((12, 6), (45, 18)) 5m = (|45 - 12| + |18 - 6|) 5 = (33 + 12) 5 = 225m;

[0162] c(A1) = 0 (passable);

[0163] b(A1): Calculate the rewards and penalties;

[0164] The distance to B7(60, 60) is 30m > R(15m), and the reward is 0;

[0165] The distance to the virtual channel (y = 60) is 30m > d3(10m), and the penalty is 16.5 (calculated according to the previous example);

[0166] The distance to the nearest boundary buoy B12(60,0) is 30m > D(5m), and the penalty is 0;

[0167] b(A1) = 0 + 16.5 + 0 = 16.5;

[0168] f(A1) = 0 + 225 + 0 + 16.5 = 241.5;

[0169] 3) Loop:

[0170] Take out A1; Expand adjacent nodes (e.g., (13,6), corresponding to (65,30));

[0171] Calculate the f value of (13,6):

[0172] g((13,6)) = g(A1) + cost(A1,(13,6)) = 0 + 5m = 5;

[0173] h((13,6)) = (|45 - 13| + |18 - 6|)5 = (32 + 12)5 = 220m;

[0174] c((13,6)) = 0;

[0175] b((13,6)): The distance to B7 is √((65 - 60) 2 + (30 - 60) 2 ) ≈ 30.4m > R, the reward is 0; The distance to the virtual channel is 30m > d3, the penalty is 16.5; The distance to B12 is 30m > D, the penalty is 0; b((13,6)) = 16.5;

[0176] f((13,6)) = 5 + 220 + 0 + 16.5 = 241.5;

[0177] Add (13,6) to the open list, and set the parent node as A1;

[0178] Continue to expand other adjacent nodes, and select the node with the minimum f value for the next expansion; The algorithm will gradually explore, avoid the net cage area, and be attracted by the low penalty area of the virtual channel (near y = 60), while considering the reward for approaching the navigation buoy, and finally find a path with the lowest comprehensive cost to reach A2;

[0179] 4) Path backtracking: Starting from the target node A2, backtrack along the parent node pointer to obtain the sequence of path points;

[0180] If it is marked as an obstacle grid when constructing the environmental map, skip these grids during the traversal of adjacent nodes in the A* algorithm (or achieve it by setting c(n) = ∞);

[0181] After path planning, the unmanned vessel periodically or on demand emits UWB (Ultra-Wideband) signals. Each navigation buoy receives the UWB signals and records the timestamps, calculates the distance differences from the unmanned vessel to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt (LM) algorithm to solve for the current position of the unmanned vessel, obtaining the TDoA positioning result;

[0182] Specifically, the unmanned vessel emits UWB signals, and N (N≥3) navigation buoys receive the UWB signals and record the timestamps t rx,i , where N is the total number of navigation buoys;

[0183] Preprocess the originally obtained timestamps t rx,i including time synchronization, antenna delay calibration, and jitter suppression to obtain high-precision timestamps t'; rx,i Among them, time synchronization ensures that the clocks of all buoys are accurately synchronized to the same time reference through PTP or a similar mechanism, which is the basis of TDoA; antenna delay calibration compensates for the fixed delays introduced by signal propagation in the transmitting antenna, receiving antenna, and related RF circuits, and these delays are usually obtained through pre-calibration; in addition, there may be random jitter in the timestamps, and the jitter effect can be reduced by digital filtering techniques or averaging multiple measurements within a short period;

[0184] Select the jth navigation buoy as the reference buoy and calculate the signal arrival time difference of other navigation buoys relative to the reference buoy, expressed as:

[0185] Δt j,i =t′ rx,i -t′ rx,j

[0186] where Δt j,i is the signal arrival time difference of the ith navigation buoy relative to the reference buoy;

[0187] Calculate the distance differences from the unmanned vessel to each navigation buoy according to the following formula:

[0188] Δd j,i =c eff ·Δt j,i

[0189] where Δd j,i is the distance difference from the unmanned vessel to the ith navigation buoy and to the reference buoy; c eff is the effective propagation speed of the signal in the propagation medium, usually approximately the speed of light c≈3×10 8 m / s;

[0190] According to the following formula, construct at least two distance difference equations to construct a nonlinear equation system:

[0191]

[0192] where (x, y) is the current position of the unmanned ship; (X i , Y i ) is the position of the i-th navigation buoy; (X j , Y j ) is the position of the reference buoy;

[0193] Each calculated distance difference Δd j,i defines a geometric locus which, in a two-dimensional plane, is a hyperbola with the reference buoy B j and buoy B i as foci. The position (x, y) of the unmanned ship must satisfy all these hyperbola equations; that is, the current position (x, y) of the unmanned ship is the intersection point of these hyperbolas. Due to the existence of measurement noise, these hyperbolas usually do not intersect precisely at one point but converge within a region; a non-linear equation system needs to be solved to find the best position estimate that fits all the distance difference measurements; at least two independent distance difference equations (i.e., at least 3 navigation buoys: 1 reference + 2 others) are required to uniquely determine the position in the 2D plane;

[0194] Since the above equation system is non-linear, the Levenberg-Marquardt algorithm is used to iteratively solve the above non-linear equation system to obtain the current position (x, y) of the unmanned ship and get the TDoA positioning result; the LM algorithm combines the advantages of the gradient descent method (stable when far from the optimal solution) and the Gauss-Newton method (fast convergence when close to the optimal solution);

[0195] After that, the unmanned ship uses the sensors it carries to collect its own state information in real time, including GPS positioning data and IMU sensor data, and fuses its own state information and the TDoA positioning result through an adaptive UKF to obtain a high-precision position estimate of the unmanned ship, thereby improving the positioning accuracy and robustness of the unmanned ship and providing a positioning quality assessment;

[0196] As Figure 6 shown, the state vector of the adaptive UKF algorithm includes position, velocity, attitude quaternion, and accelerometer bias, and through iterative calculations in the prediction step and the update step, the optimal estimation of the state of the unmanned ship is achieved;

[0197] State vector: x k = [px, py, pz, vx, vy, vz, qw, qx, qy, qz, bx, by, bz, bgx, bgy, bgz] T ;

[0198] where px, py, pz are the positions (in meters) of the unmanned ship in the global coordinate system; vx, vy, vz are the velocities (in m / s) of the unmanned ship in the global coordinate system; qw, qx, qy, qz are the attitude quaternions of the unmanned ship (dimensionless); bx, by, bz are the biases of the accelerometer (in m / s²); bgx, bgy, bgz are the biases of the gyroscope (in rad / s);

[0199] Control input: u k = [ax_m, ay_m, az_m, wx_m, wy_m, wz_m] T ;

[0200] where ax_m, ay_m, az_m are the accelerations measured by the IMU (in m / s²); wx_m, wy_m, wz_m are the angular velocities measured by the IMU (in rad / s);

[0201] In the prediction step, based on the IMU sensor data at the current moment, the state estimation of the unmanned ship at the next moment is predicted using the adaptive UKF algorithm, expressed as:

[0202]

[0203] where, is the state vector at the predicted k-th moment; f is the state transition function, which predicts the current state based on the previous state and the control input; x k-1 is the state vector at the (k - 1)-th moment; w k is the process noise;

[0204] The state transition function f includes:

[0205] a) Position update:

[0206] p_k = p_(k - 1) + v_(k - 1) * Δt + 0.5 * (R(q_(k - 1)) * (a_m - b_a_(k - 1)) + g) * Δt 2 [[ID=�7]]

[0207] where p_k is the position vector at the k-th moment; p_(k - 1) is the position vector at the (k - 1)-th moment; v_(k - 1) is the velocity vector at the (k - 1)-th moment; Δt is the time interval; R(q_(k - 1)) is the rotation matrix corresponding to the attitude quaternion at the (k - 1)-th moment; a_m is the acceleration vector measured by the IMU; b_a_(k - 1) is the accelerometer bias vector at the (k - 1)-th moment; g is the gravitational acceleration vector;

[0208] b) Velocity update:

[0209] v_k = v_(k - 1) + (R(q_(k - 1)) * (a_m - b_a_(k - 1)) + g) * Δt

[0210] where: \(v_k\) is the velocity vector at time \(k\);

[0211] c) Attitude update (quaternion):

[0212]

[0213] where \(q_k\) is the attitude quaternion at time \(k\); \(q_{(k - 1)}\) is the attitude quaternion at time \(k - 1\); is quaternion multiplication; \(quat\_update(w_m - b_g_{(k - 1)}, \Delta t)\) is the quaternion updated according to the angular velocity change;

[0214] d) Bias update:

[0215] \(b_{a_k}=b_{a_{(k - 1)}}\), \(b_{g_k}=b_{g_{(k - 1)}}\) (assuming the bias changes slowly)

[0216] where \(b_{a_k}\) is the accelerometer bias vector at time \(k\); \(b_{g_k}\) is the gyroscope bias vector at time \(k\);

[0217] Sigma point generation: UKF uses Sigma points to approximate the state distribution instead of linearizing the non - linear function like EKF; Calculate the number of Sigma points: \(2L + 1\), where \(L\) is the dimension of the state vector (16 in this example); Calculate the Sigma points: \(X^{(0)}_{(k - 1)}=x_{(k - 1)}\), \(X^{(i)}_{(k - 1)}=x_{(k - 1)}+(\sqrt{(L+\lambda)*P_{(k - 1)}})_i\), \(i = 1,\cdots,L\), \(X^{(i + L)}_{(k - 1)}=x_{(k - 1)}-(\sqrt{(L+\lambda)*P_{(k - 1)}})_i\), \(i = 1,\cdots,L\);

[0218] where \(X^{(0)}_{(k - 1)}\) is the 0th Sigma point, equal to the state estimate at time \(k - 1\); \(X^{(i)}_{(k - 1)}\) is the \(i\)th Sigma point; \(x_{(k - 1)}\) is the state estimate at time \(k - 1\); \(P_{(k - 1)}\) is the covariance matrix at time \(k - 1\); \(\lambda\) is a scaling factor used to adjust the distribution of Sigma points; usually set as \(\lambda=\alpha\) 2 \(*(L+\kappa)-L\), where \(\alpha\) controls the degree of diffusion of Sigma points in the state distribution (usually set as a small positive number, such as 0.001), \(\kappa\) is a secondary scaling factor (usually set as 0); \((\sqrt{(L+\lambda)*P_{(k - 1)}})_i\) is the \(i\)th column of the square root of the matrix \((L+\lambda)*P_{(k - 1)}\); methods such as Cholesky decomposition can be used to calculate the square root of the matrix;

[0219] Sigma point propagation: Propagate the Sigma points through the kinematic model to obtain the predicted Sigma points:

[0220]

[0221] Wherein, is the i-th predicted Sigma point; f(X^(i)_(k-1),uk) is the state transition function that propagates the i-th Sigma point to time k;

[0222] Predicted state estimation and covariance estimation: Calculate the predicted state estimation and covariance estimation based on the propagated Sigma points:

[0223]

[0224] Wherein, is the predicted state estimation at time k; is the predicted covariance estimation at time k; W^(m)_i is the state estimation weight of the i-th Sigma point; W^(c)_i is the covariance estimation weight of the i-th Sigma point; Qk is the process noise covariance matrix; The calculation formula for the weights is:

[0225] W^(m)_0 = λ / (L + λ)

[0226] W^(c)_0 = λ / (L + λ) + (1 - α 2 + β)

[0227] W^(m)_i = W^(c)_i = 1 / (2*(L + λ)), i = 1,..., 2L

[0228] Wherein, λ is the scaling factor; α is a UKF parameter that controls the spread of Sigma points in the state distribution; β is a UKF parameter that is a factor for incorporating prior knowledge of the state distribution; For a Gaussian distribution, it is usually set to 2;

[0229] In the update step, calculate the residuals of the TDoA positioning result and the GPS positioning data respectively, further calculate the weighted sum of squares of the adaptive weights of the two residuals, and finally perform fusion update, and use the updated state estimation as the high-precision position estimation of the unmanned ship;

[0230] Specifically, calculate the residuals:

[0231] TDoA residual:

[0232]

[0233] Wherein, yk_tdoa is the TDoA residual vector; zk_tdoa is the actual TDoA measurement value vector; is the predicted TDoA measurement vector (obtained by passing the Sigma points into the observation space);

[0234] GPS residual:

[0235]

[0236] where yk_gps is the GPS residual vector; zk_gps is the actual GPS measurement vector; is the predicted GPS measurement vector (obtained by passing the Sigma points into the observation space);

[0237] Calculate the weighted sum of squared residuals (WSSR):

[0238] TDoA WSSR:

[0239] WSSR_tdoa = yk_tdoa^T * P_zz_tdoa^(-1) * yk_tdoa

[0240] where WSSR_tdoa is the weighted sum of squared residuals of TDoA; P_zz_tdoa is the covariance matrix of TDoA predicted observations;

[0241] GPS WSSR:

[0242] WSSR_gps = yk_gps^T * P_zz_gps^(-1) * yk_gps

[0243] where WSSR_gps is the weighted sum of squared residuals of GPS; P_zz_gps is the covariance matrix of GPS predicted observations;

[0244] WSSR reflects the degree of difference between the measurement value and the predicted value; the larger the WSSR, the lower the credibility of the measurement value;

[0245] Calculate the adaptive weight:

[0246] TDoA weight:

[0247] weight_tdoa = exp(-WSSR_tdoa / (2 * threshold_tdoa))

[0248] where weight_tdoa is the weight of TDoA; threshold_tdoa is a preset threshold for controlling the change speed of the TDoA weight; G

[0249] PS weight:

[0250] weight_gps = exp(-WSSR_gps / (2 * threshold_gps))

[0251] Among them, weight_gps is the weight of GPS; threshold_gps is a pre-set threshold for controlling the change speed of the GPS weight;

[0252] Weight range: The weight value is between 0 and 1. The larger the WSSR, the smaller the weight;

[0253] Fusion update:

[0254] Calculate the weighted residual:

[0255] yk = weight_tdoa * yk_tdoa + weight_gps * yk_gps

[0256] Among them, yk is the weighted residual vector;

[0257] If only TDoA or GPS data is available, only the corresponding data is used for update;

[0258] Calculate the weighted observation covariance matrix:

[0259] P_zz = weight_tdoa * P_zz_tdoa + weight_gps * P_zz_gps

[0260] Among them, P_zz is the weighted observation covariance matrix;

[0261] Calculate the weighted cross-covariance matrix:

[0262] P_xz = weight_tdoa * P_xz_tdoa + weight_gps * P_xz_gps

[0263] Among them, P_xz is the weighted cross-covariance matrix;

[0264] Calculate the Kalman gain:

[0265] Kk = P_xz * P_zz^(-1)

[0266] Among them, Kk is the Kalman gain;

[0267] Update the state estimate: [[ID=5B]]

[0268]

[0269] Among them, [[ID=ID=59]] is the state estimate at the updated k-th moment;

[0270] Update the covariance estimate:

[0271]

[0272] where \(P_k\) is the covariance estimate at time \(k\) after update;

[0273] This embodiment can also use buoys for calibration. When the unmanned ship approaches the navigation buoy, the TDoA positioning accuracy is usually higher; this can be used to enhance the calibration effect of the adaptive UKF; the calibration method is as follows:

[0274] a) Dynamically adjust the observation noise \(R_k\): When the distance between the unmanned ship and the navigation buoy participating in TDoA positioning is relatively close (e.g., judged by the position estimated by the adaptive UKF), reduce the variance of the corresponding TDoA measurement value in \(R_k\); this will make the adaptive UKF trust the TDoA measurement value more in the update step, thus more effectively correcting the position estimate;

[0275] b) Trigger more frequent TDoA positioning: When approaching the buoy, the UWB signal transmission frequency or the TDoA solution frequency can be increased to obtain more high-precision positioning information for adaptive UKF update;

[0276] c) Directly reset the state: In very few cases, if it is certain that the TDoA positioning result is very reliable (e.g., multiple buoys are close, and the GDOP value is extremely low), it can be considered to directly correct (reset) the position state of the adaptive UKF with the TDoA positioning result; however, this may introduce discontinuity and needs to be carefully evaluated;

[0277] d) Adjust the adaptive weights: When approaching the buoy, the adaptive weight of TDoA can be appropriately increased, and the adaptive weight of GPS can be reduced (if the GPS signal quality is poor);

[0278] Finally, the unmanned ship performs an autonomous inspection task according to the global reference path and sends the inspection data to the control center on the shore in real time; LoRaWAN can be selected for long-distance and low-power communication between the unmanned ship and the control center, or combined with WiFi / 4G / 5G to achieve high-bandwidth data transmission with the control center;

[0279] During the inspection process, based on the deviation (including lateral deviation and heading deviation) between the high-precision position estimate of the unmanned ship (from the adaptive UKF fusion positioning result) and the global reference path, the heading of the unmanned ship is calibrated in real time based on a PID controller to achieve precise path tracking and calibration; when the unmanned ship approaches the navigation buoy, the observation noise of the TDoA measurement value in the adaptive UKF is dynamically adjusted, the weight of the TDoA positioning result is increased for calibration, and the positioning accuracy is further improved;

[0280] Specifically, as Figure 7 shown, search for the point on the global reference path that is closest to the current position \(p\) of the unmanned ship kThe nearest point p ref ; Calculate the current position p of the unmanned ship k to the point p ref The perpendicular distance to the tangent of the path where it is located is used as the lateral deviation e d ; Calculate the current heading ψ of the unmanned ship k and the point p ref The angle difference between the tangent direction ψ of the path where it is located ref is used as the heading deviation e ψ ;

[0281] Input the lateral deviation e d and the heading deviation e ψ into the PID controller together. The output of the PID controller is the control command, which includes at least the desired rudder angle δ cmd ; The control law of the PID controller is expressed as:

[0282] δ cmd = delta d + delta ψ

[0283]

[0284] where delta d represents the lateral deviation control amount; delta ψ represents the heading deviation control amount; K p,d , K i,d and K d,d represent the proportional gain, integral gain and derivative gain of the lateral deviation control respectively; K p,ψ and K d,ψ represent the proportional gain and derivative gain of the heading deviation control;

[0285] In addition, based on the positioning uncertainty or adaptive weight of the adaptive UKF, adaptively adjust the gain parameters of the PID controller;

[0286] 1) Adaptive adjustment of the gain parameters of the PID controller based on the positioning uncertainty of the adaptive UKF includes: calculating the positioning uncertainty index using the covariance matrix provided by the adaptive UKF algorithm. When the positioning uncertainty index is greater than the preset uncertainty threshold, reduce the gain parameters of the PID controller to reduce the sensitivity to noise; otherwise, increase the gain parameters of the PID controller to improve the control accuracy;

[0287] 2) Adaptive adjustment of the gain parameters of the PID controller based on adaptive weights includes: obtaining the adaptive weights of the TDoA positioning results and GPS positioning data. When any one of the adaptive weights is less than the preset weight threshold, the gain parameters of the PID controller are reduced to reduce the dependence on unreliable sensor data; otherwise, the gain parameters of the PID controller are increased to better utilize the high-precision positioning results.

[0288] In addition, a safety mechanism is also set up in this embodiment. When abnormal situations such as system failures, communication interruptions, unreliable positioning results, detection of obstacles, approaching the boundary, and low battery occur during the inspection process, the unmanned boat triggers actions such as decelerating and hovering, returning, or dynamic obstacle avoidance, and sends an alarm to the control center, waiting for manual intervention.

[0289] The unmanned boat sails autonomously according to the above steps, executes the cage inspection task, and at the same time conducts information interaction with the intelligent buoy network, obtains environmental data and boundary early warning information and sends them to the control center in real time, so as to achieve safe and efficient autonomous inspection.

[0290] This method can achieve autonomous, safe, and efficient inspection of unmanned boats in dense cage areas, improve the intelligent level of aquaculture management, and is especially suitable for application scenarios with high positioning accuracy requirements.

[0291] The same or similar reference numerals correspond to the same or similar components.

[0292] The terms describing the positional relationship in the drawings are only for illustrative purposes and should not be construed as a limitation of this application.

[0293] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. An intelligent inspection method for unmanned vessels of ocean cages based on buoy networks and TDoA positioning, characterized in that, It includes the following steps: S1: Construct an intelligent buoy network in the marine cage aquaculture area, including several navigation buoys and several boundary buoys; the navigation buoys are used to provide UWB TDoA positioning reference and virtual navigation channels for the unmanned ship; the boundary buoys are used to identify the boundary of the aquaculture area; S2: Based on the intelligent buoy network, construct a refined grid map of the aquaculture area, and divide the aquaculture area into passable areas, cage no-go areas, buoy no-go areas, and areas outside the boundary; S3: Based on the refined environment map, use an improved A* algorithm for global path planning to generate a global reference path; the cost function of the improved A* algorithm introduces additional costs based on grid types and buoy-related costs; S4: The unmanned ship emits UWB signals, each navigation buoy receives the UWB signals and records timestamps, calculates the distance differences from the unmanned ship to each navigation buoy based on the TDoA principle, and uses the Levenberg-Marquardt algorithm to solve the current position of the unmanned ship to obtain the TDoA positioning result; S5: The unmanned ship uses the sensors carried by itself to collect its own state information in real time, and fuses its own state information and the TDoA positioning result through an adaptive UKF to obtain a high-precision position estimate of the unmanned ship; S6: The unmanned ship executes an autonomous inspection task according to the global reference path and sends the inspection data to the control center on the shore in real time. During the inspection process, based on the deviation between the high-precision position estimate of the unmanned ship and the global reference path, the heading of the unmanned ship is calibrated in real time based on a PID controller to achieve precise path tracking and calibration.

2. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 1, characterized in that, In the step S1, the navigation buoy is equipped with a UWB transceiver module, a GPS positioning module, a LoRa module, and an environmental monitoring sensor, and uses the IEEE 1588 Precision Time Protocol for clock synchronization, and is used to provide UWB TDoA positioning reference, virtual navigation channels, and environmental monitoring data for the unmanned ship; The boundary buoy is equipped with a UWB transceiver module, a GPS positioning module, a LoRa module, and an environmental monitoring sensor, and is used to identify the boundary of the aquaculture area and provide environmental monitoring data; The unmanned ship is equipped with a UWB transceiver module, a GPS positioning module, an IMU sensor, a controller, an adaptive UKF multi-sensor fusion module, a PID path tracking module, a map construction module, and a communication module, and is used to execute an autonomous inspection task and achieve information interaction, high-precision positioning, and precise path tracking and calibration.

3. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 2, characterized in that, The navigation buoy also obtains a rough position using the GPS positioning module, and performs cooperative positioning using the least squares method and the RANSAC algorithm through UWB TDoA measurements with surrounding navigation buoys and boundary buoys with known positions, and continuously optimizes its own position estimate.

4. The intelligent inspection method for an unmanned boat of an ocean cage based on a buoy network and TDoA positioning according to claim 1, wherein, In the step S2, the divided areas of the refined grid map include: Occupied cage no-go area: including all grids within a preset range around the occupied cages; Vacant cage no-go area: including all grids within a preset range around the vacant cages; Buoy no-go area: including all grids within a certain range around the navigation buoy or the boundary buoy; Off - boundary area: includes all grids located outside the effective range of the aquaculture area defined by boundary buoys; Passable area: includes all other grids except the above - mentioned no - go area and off - boundary area.

5. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 1, wherein In the step S3, the cost function of the improved A* algorithm includes: f(n) = g(n)+h(n)+c(n)+b(n) b(n) = b reward (n) + b penalty (n) + b boundary (n) Among them, f(n) is the total cost of moving from the starting point to the current nth grid node; g(n) is the actual moving cost of moving from the starting point to the current nth grid node; h(n) is the heuristic estimated cost of moving from the current nth grid node to the target grid node; c(n) is the additional cost based on the grid type. For grids in no-go areas and outside the boundary, c(n) = ∞, and for passable areas, c(n) = 0; b(n) is the buoy-related cost, b reward (n) is the reward for approaching the navigation buoy, b penalty (n) is the penalty for deviating from the center line of the virtual navigation channel; b boundary (n) is the penalty for moving away from the boundary buoy; Said b reward (n) is expressed as: b reward (n) = -k·max(0, R - d) where d is the distance from the current n - th grid node to the nearest navigation buoy; R is the influence range radius of the navigation buoy; k is the reward coefficient; Said b penalty (n) is expressed as: where d channel is the vertical distance from the current nth grid node to the centerline of the nearest virtual navigation channel; d1, d2, and d3 are the first, second, and third distance thresholds respectively, used to divide different degrees of deviation; α1, α2, and α3 are the first, second, and third penalty coefficients respectively, and α1 < α2 < α3, used to control the penalty strength of different degrees of deviation to ensure that the greater the deviation, the greater the cost; The said b boundary (n) is expressed as: b boundary (n) = β·max(0, D - d bound ) where d bound is the distance from the current nth grid node to the nearest boundary buoy, β is the penalty coefficient, and D is the boundary buoy safety distance threshold.

6. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 1, characterized in that In the step S4, the unmanned ship emits a UWB signal, and N navigation buoys receive the UWB signal and record the timestamp t rx,i , where N is the total number of navigation buoys; Preprocess the originally obtained timestamp t rx,i including time synchronization, antenna delay calibration, and jitter suppression to obtain a high-precision timestamp t' rx,i ; Select the j - th navigation buoy as the reference buoy, and calculate the time difference of arrival of signals of other navigation buoys relative to the reference buoy, expressed as: Δt j,i = t' rx,i - t' rx,j where Δt j,i is the signal arrival time difference of the i-th navigation buoy relative to the reference buoy; Calculate the distance difference between the unmanned ship and each navigation buoy according to the following formula: Δd j,i = c eff ·Δt j,i where, Δd j,i is the distance difference between the unmanned ship and the i-th navigation buoy and the reference buoy; c eff is the effective propagation speed of the signal in the propagation medium; According to the following formula, construct at least two distance - difference equations to construct a non - linear equation system: Among them, (x, y) is the current position of the unmanned ship; (X i , Y i ) is the position of the i-th navigation buoy; (X j , Y j ) is the position of the reference buoy; Use the Levenberg - Marquardt algorithm to iteratively solve the non - linear equation system to obtain the current position (x, y) of the unmanned ship and get the TDoA positioning result.

7. An intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 1, characterized in that, In the step S5, the self - state information collected in real - time by the positioning sensors carried by the unmanned ship at least includes: GPS positioning data and IMU sensor data; Fuse the self - state information and the TDoA positioning result through an adaptive UKF, including a prediction step and an update step executed in sequence; In the prediction step, based on the IMU sensor data at the current moment, predict the state estimate of the unmanned ship at the next moment based on the adaptive UKF algorithm; in the update step, calculate the residuals of the TDoA positioning result and the GPS positioning data respectively, further calculate the weighted sum of squares of the two residuals and the adaptive weight, and finally perform fusion update, and use the updated state estimate as the high - precision position estimate of the unmanned ship.

8. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 7, characterized in that, In the step S6, based on the deviation between the high - precision position estimate of the unmanned ship and the global reference path, calibrate the heading of the unmanned ship in real - time based on a PID controller, including the following steps: Search for the point p k nearest to the current position p of the unmanned ship ref on the global reference path; calculate the current position p k of the unmanned ship to the point p ref on the path, and take the perpendicular distance to the tangent of the path as the lateral deviation e d ; calculate the current heading ψ k of the unmanned ship and the tangent direction ψ ref of the path where the point p ref is located, and take the angular difference as the heading deviation e ψ ; Input the lateral deviation e d and the heading deviation e ψ jointly into a PID controller, and the output of the PID controller is a control command, which at least includes the desired rudder angle δ cmd ; the control law of the PID controller is expressed as: δ cmd = delta d + delta ψ Among them, delta d represents the lateral deviation control amount; delta ψ represents the course deviation control amount; K p,d , K i,d and K d,d respectively represent the proportional gain, integral gain and derivative gain of lateral deviation control; K p,ψ and K d,ψ represent the proportional gain and derivative gain of course deviation control.

9. The intelligent inspection method for an unmanned ship of an ocean cage based on a buoy network and TDoA positioning according to claim 8, characterized in that, The step S6 also includes: adaptively adjusting the gain parameters of the PID controller based on the positioning uncertainty or adaptive weight of the adaptive UKF; Adaptive adjustment of the gain parameters of the PID controller based on the positioning uncertainty of the adaptive UKF includes: calculating the positioning uncertainty index using the covariance matrix provided by the adaptive UKF algorithm, and when the positioning uncertainty index is greater than the preset uncertainty threshold, reducing the gain parameters of the PID controller; otherwise, increasing the gain parameters of the PID controller; Adaptive adjustment of the gain parameters of the PID controller based on the adaptive weight includes: obtaining the adaptive weights of the TDoA positioning result and the GPS positioning data, and when any one of the adaptive weights is less than the preset weight threshold, reducing the gain parameters of the PID controller; otherwise, increasing the gain parameters of the PID controller.

10. A method for intelligent inspection of unmanned vessels for ocean cages based on buoy networks and TDoA positioning according to any one of claims 1 to 9, characterized in that, The method also includes: when an abnormal situation occurs during the inspection, the unmanned ship triggers deceleration and hovering, returning or dynamic obstacle avoidance actions, and sends an alarm to the control center and waits for manual intervention; The abnormal situation at least includes any one of system failure, communication interruption, unreliable positioning result, detected obstacle, approaching the boundary, and low battery.

Images