Path planning method, system and device for unmanned ship and storage medium

Abstract

The invention discloses a path planning method, system and device for an unmanned ship and a storage medium, and the method comprises the steps: 1, collecting the environmental parameters of an area in a pond, constructing a global map, and inputting the geographic coordinates of a task area; setting bait casting point distribution and adjusting navigation parameters; 2, acquiring a current position, delimiting a safe navigation boundary, and ensuring that subsequent movement does not exceed a preset range; 3, carrying out route calculation based on an algorithm fusing a dynamic weight mechanism and an adaptive cost function, and generating a global route plan; according to global path planning, an overall driving route from a starting point to a target point is generated based on known environment information, the local path module senses the surrounding environment in real time depending on a sensor carried by the unmanned ship, dynamic data such as the obstacle position, the water flow direction and the speed are obtained, obstacle avoidance and course correction are achieved through rapid re-planning or behavior decision making, and the unmanned ship is more intelligent. According to the mechanism, the calculation overhead caused by complete re-planning is avoided, and the problem that the local part falls into the local minimum is also prevented.

Description

Technical Field

[0001] This invention belongs to the field of unmanned surface vessel (USV) technology and relates to a path planning method, system, device, and storage medium for USVs. Background Technology

[0002] Currently, by equipping navigation and positioning modules, sensor systems, and automatic control devices, unmanned surface vessels (USVs) can autonomously navigate along preset paths and complete targeted and quantitative feeding tasks, significantly improving operational efficiency and management precision. However, existing path planning algorithms are mostly derived from land robots or aerial drones, and their direct application to USVs faces challenges in adaptability. For example, static environment assumptions struggle to cope with dynamic obstacle interference, global planning lacks real-time obstacle avoidance mechanisms, and most algorithms do not fully consider the USV's own nonholonomic constraints and low-speed maneuvering characteristics. Furthermore, aquaculture ponds often experience overgrown aquatic plants, scattered buoys, and crisscrossing nets, further increasing the difficulty of determining path reachability and trajectory smoothness. Summary of the Invention

[0003] In order to overcome at least one deficiency of the prior art, the present invention provides a path planning method, system, device and storage medium for unmanned vessels.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a path planning method for an unmanned surface vessel, comprising the following steps: Step 1: Collect environmental parameters of the pond area, construct a global map, and enter the geographical coordinates of the task area; set the distribution of feeding points and adjust navigation parameters; Step 2: Obtain the current position, define the safe navigation boundary, and ensure that subsequent movement does not exceed the preset range; Step 3: Calculate the flight path based on an algorithm that integrates a dynamic weighting mechanism and an adaptive cost function to generate a global route plan; Step 4: Decompose the global path into several segments and calculate the target heading angle and propulsion speed of the current segment in real time; Step 5: Continuously collect environmental data and input the data into the local obstacle avoidance module. If a dynamic obstacle is detected, replan the local trajectory and incrementally update the global path. Environmental data includes changes in water depth, location of floating objects, and disturbances from wind and waves; Step 6: Regularly upload the location status and equipment health indicators of the unmanned vessel; when it reaches the designated feeding point, activate the feeding device, release feed according to the preset dosage, and record the feeding time and coordinate information for later traceability; Step 7: After feeding at the current feeding point, determine if there are still unfinished tasks. If yes, continue to the next feeding point; otherwise, initiate the return procedure and return to the starting position along the shortest safe path.

[0005] Furthermore, the global path planning method is as follows: Step 31: Build a hybrid hierarchical model and pre-generate a rough boot path; Step 32: Dynamically adjust the coarse guiding path based on environmental uncertainty factors to generate a global path plan; Step 33: Use a genetic algorithm to determine the order of visiting feeding points.

[0006] Furthermore, the method in step 31 is as follows: Based on the hybrid modeling method of raster map and topology map, a hybrid hierarchical model is constructed. The bottom layer of the model uses a raster map to describe the pond terrain and obstacle locations; the model extracts key landmarks of the navigable area to form a simplified topology map. The IA* algorithm runs on the bottom layer grid, and the Dijkstra algorithm is used to pre-generate a coarse guiding path at the top layer.

[0007] Furthermore, the method in step 32 is as follows: Let the current node be "n", and its total cost function be defined as follows: f(n) = g(n) + \omega(n) \cdot h(n); Where "g(n)" represents the actual path cost from the starting point to node "n"; "h(n)" is the heuristic value for Euclidean or Manhattan distance estimation; "\omega(n)" is a dynamic weighting coefficient that varies with environmental conditions, and "\omega(n)" is adjusted online based on the local obstacle density "\rho(n)" and the water flow velocity vector "\vec{v}_w(n)", as shown in the formula. \omega(n) = 1 + \alpha \cdot \frac{\rho(n)}{\rho_{max}} + \beta \cdot\|\vec{v}_w(n)\| Among them, "\alpha" and "\beta" are empirical parameter tuning coefficients that control the sensitivity to the influence of obstacle density and water flow, respectively, with a value range of [0.5, 1.2]. When there are dense obstacles or strong water flow in a certain area, the heuristic weight of the nodes in that area is automatically increased.

[0008] Furthermore, the method of step 33 is as follows: Path planning is decomposed into two stages: sequence optimization and single-segment path generation. Phase 1 Sequence Optimization: Constructing a Traveling Salesman Problem with Time Windows (TSP-TW), and using a genetic algorithm to find the minimum total travel distance and the optimal visiting order. Chromosome coding uses an integer arrangement, with each gene representing a feeding point number. The fitness function is defined as follows: F(\pi) = \frac{1}{C(\pi) + \gamma \cdot T_{violation}} Where "C(\pi)" is the total path length, "T_{violation}" is the cumulative duration of violating the time window constraint, "\gamma" is the penalty coefficient, the crossover operation adopts sequential crossover (OX), and the mutation adopts reverse mutation to maintain the validity of the solution. After multiple generations of evolution, the optimal access sequence "\pi^*" is output as the input for subsequent segmented path planning; The second stage involves single-segment path generation: based on this sequence, piecewise IA* planning is performed, and path smoothing is introduced. Post-processing is performed using a combination of cubic spline interpolation and Bézier curve fitting. For any three consecutive path points "P_{i-1}, P_i, P_{i+1}", a quadratic Bézier curve is constructed. B(t) = (1-t)^2 P_{i-1} + 2t(1-t)P_c + t^2 P_{i+1}, \quad t \in [0,1] The control point "P_c" is set to "(P_{i-1} + P_{i+1}) / 2" to ensure curvature continuity. At the same time, it checks whether each curve segment crosses the obstacle buffer. If there is a conflict, an intermediate waypoint is inserted for refitting.

[0009] Furthermore, the local obstacle avoidance module in step 5 adopts a dynamic window algorithm. Dynamic obstacles are sudden obstacles. When a dynamic obstacle causes the current unmanned vessel path to be interrupted, the local trajectory is replanned.

[0010] Furthermore, the method in step 5 is as follows: When the unmanned vessel approaches an obstacle on the pre-planned path and the distance is less than the threshold "d_{safe} = 2m", a local replanning mechanism is triggered. DWA samples several feasible motion commands in the velocity space "(v, \omega)" and selects the optimal instantaneous action based on the evaluation function: U(v, \omega) = w_1 \cdot \text{heading}(v, \omega) + w_2 \cdot \text{dist}(v, \omega) + w_3 \cdot \text{vel}(v, \omega) Among them, the heading term measures the angular deviation of the forward direction from the target, the dist term reflects the distance to the nearest obstacle, the vel term encourages maintaining a high linear speed, and the weights "w_i" are dynamically adjusted according to the task stage: in open water, "w_3" is increased to improve speed, and in narrow passages, "w_2" is enhanced to ensure safety.

[0011] A path planning device for an unmanned surface vessel, comprising: The initial adjustment module is used to collect environmental parameters within the pond area, construct a global map, input the geographical coordinates of the task area, set the distribution of feeding points, and adjust navigation parameters. The global route planning module calculates the route based on an algorithm that integrates a dynamic weighting mechanism and an adaptive cost function, generating a global route plan. The local obstacle avoidance module is used to replan local trajectories; The feeding module is used to periodically upload the location status and health indicators of the unmanned vessel; when it reaches the designated feeding point, it activates the feeding device, releases feed according to the preset dosage, and records the feeding time and coordinate information for later traceability.

[0012] A path planning system for an unmanned surface vessel (USV) includes a memory and a processor. The memory stores a computer program that is executed by the processor. When the computer program is run by the processor, it executes the path planning method for the USV.

[0013] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the path planning method for the unmanned vessel.

[0014] In summary, the advantages of this invention are: The global path planning in this application generates an overall driving route from the starting point to the target point based on known environmental information. The local path module relies on the sensors carried by the unmanned vessel to perceive the surrounding environment in real time, and obtain dynamic data such as obstacle position, water flow direction and speed. Obstacle avoidance and course correction are achieved through rapid replanning or behavioral decision-making. This mechanism avoids the computational overhead caused by complete replanning and also prevents the problem of getting stuck in local minima. Attached Figure Description

[0015] Figure 1 This is a flowchart of the path planning method of the present invention. Detailed Implementation

[0016] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0017] Example: like Figure 1 As shown, a path planning method for an unmanned surface vessel includes the following steps: Step 1: Collect environmental parameters of the pond area, construct a global map, and enter the geographical coordinates of the task area; set the distribution of feeding points and adjust navigation parameters; Step 2: Obtain the current position, define the safe navigation boundary, and ensure that subsequent movement does not exceed the preset range; Step 3: Calculate the flight path based on an algorithm that integrates a dynamic weighting mechanism and an adaptive cost function to generate a global route plan; Step 4: Decompose the global path into several segments and calculate the target heading angle and propulsion speed of the current segment in real time; Step 5: Continuously collect environmental data and input the data into the local obstacle avoidance module. If a dynamic obstacle is detected, replan the local trajectory and incrementally update the global path. Environmental data includes changes in water depth, location of floating objects, and disturbances from wind and waves; Step 6: Regularly upload the location status and equipment health indicators of the unmanned vessel; when it reaches the designated feeding point, activate the feeding device, release feed according to the preset dosage, and record the feeding time and coordinate information for later traceability; Step 7: After feeding at the current feeding point, determine if there are still unfinished tasks. If yes, continue to the next feeding point; otherwise, initiate the return procedure and return to the starting position along the shortest safe path.

[0018] The global path planning method is as follows: Step 31: Build a hybrid hierarchical model and pre-generate a rough boot path; A hybrid hierarchical model is constructed based on a hybrid modeling approach combining raster maps and topology graphs. The bottom layer uses a high-resolution raster map to describe the pond's topography and obstacle locations; the model extracts key landmarks (including the center of the feeding area and turning nodes) from the navigable area to form a simplified topology graph. The IA* algorithm runs on the bottom layer mesh, while the top layer uses Dijkstra's algorithm to pre-generate a coarse guiding path as a directional constraint for IA*, thereby reducing the number of invalid expanded nodes.

[0019] Step 32: Dynamically adjust the coarse guiding path based on environmental uncertainty factors to generate a global path plan; Let the current node be "n", and its total cost function be defined as follows: f(n) = g(n) + \omega(n) \cdot h(n); Where "g(n)" represents the actual path cost from the starting point to node "n"; "h(n)" is the heuristic value for Euclidean or Manhattan distance estimation; "\omega(n)" is a dynamic weighting coefficient that varies with environmental conditions, and "\omega(n)" is adjusted online based on the local obstacle density "\rho(n)" and the water flow velocity vector "\vec{v}_w(n)", as shown in the formula. \omega(n) = 1 + \alpha \cdot \frac{\rho(n)}{\rho_{max}} + \beta \cdot\|\vec{v}_w(n)\| Here, "\alpha" and "\beta" are empirical parameter tuning coefficients that control the sensitivity to obstacle density and water flow influence, respectively, with values ​​ranging from [0.5, 1.2]. When there are dense obstacles or strong water flow in a certain area, the heuristic weight of the nodes in that area is automatically increased, prompting the algorithm to prioritize exploring safer and lower-energy alternative paths.

[0020] Nodes include feeding points, turning points, etc.

[0021] The dynamic adjustments in this step are based on the density of static obstacles in different areas and the current water flow velocity, which differs from the subsequent local obstacle avoidance.

[0022] Step 33: Use a genetic algorithm to determine the order in which feeding points are visited; Path planning is decomposed into two stages: sequence optimization and single-segment path generation. The first stage is sequence optimization: A Traveling Salesman Problem (TSP-TW) with time windows is constructed, and a genetic algorithm is used to solve for the minimum total travel distance and the optimal visiting order. Specifically: Chromosome coding uses an integer arrangement, with each gene representing a feeding point number. The fitness function is defined as follows: F(\pi) = \frac{1}{C(\pi) + \gamma \cdot T_{violation}} Where "C(\pi)" represents the total path length, "T_{violation}" represents the cumulative duration of violations of the time window constraint, and "\gamma" represents the penalty coefficient. Sequential crossover (OX) is used for crossover operations, and reverse mutation is used to maintain the validity of the solution. After multiple generations of evolution, the optimal access sequence "\pi^*" is output as the input for subsequent segmented path planning.

[0023] The second stage involves single-segment path generation: Piecewise IA* planning is performed based on this sequence, and path smoothing is introduced. Post-processing is performed using a combination of cubic spline interpolation and Bézier curve fitting. For any three consecutive path points "P_{i-1}, P_i, P_{i+1}", a quadratic Bézier curve is constructed: B(t) = (1-t)^2 P_{i-1} + 2t(1-t)P_c + t^2 P_{i+1}, \quad t \in [0,1] The control point "P_c" is set to "(P_{i-1} + P_{i+1}) / 2" to ensure curvature continuity. Simultaneously, each curve segment is checked for crossing obstacle buffer zones; if a conflict exists, an intermediate waypoint is inserted for refitting.

[0024] Step 5's local obstacle avoidance module employs a dynamic window algorithm. Dynamic obstacles are sudden obstacles (such as floating objects, other vessels, etc.). When a dynamic obstacle causes the current unmanned vessel's path to be interrupted, the local trajectory is replanned, specifically including... When an unmanned vessel approaches an obstacle on a pre-planned path and the distance is less than the threshold "d_{safe} = 2m", a local replanning mechanism is triggered.

[0025] DWA samples several feasible motion commands in the velocity space "(v, \omega)" and selects the optimal instantaneous action based on the evaluation function: U(v, \omega) = w_1 \cdot \text{heading}(v, \omega) + w_2 \cdot \text{dist}(v, \omega) + w_3 \cdot \text{vel}(v, \omega) Among them, the heading term measures the angular deviation of the forward direction from the target, the dist term reflects the distance to the nearest obstacle, and the vel term encourages maintaining a high linear velocity. The weights "w_i" are dynamically adjusted according to the mission phase: in open water, "w_3" is increased to improve speed, and in narrow passages, "w_2" is enhanced to ensure safety.

[0026] The global path planning in this application generates an overall driving route from the starting point to the target point based on known environmental information. The local path module relies on the sensors carried by the unmanned vessel to perceive the surrounding environment in real time, and obtain dynamic data such as obstacle position, water flow direction and speed. Obstacle avoidance and course correction are achieved through rapid replanning or behavioral decision-making. This mechanism avoids the computational overhead caused by complete replanning and also prevents the problem of getting stuck in local minima.

[0027] Based on the actual geographical features and operational requirements of mud crab farming ponds, a high-fidelity virtual water environment scene was constructed. The environmental model employs a raster map representation, dividing the farming area into several discrete units. Each unit is labeled with its topographical attributes, including water depth, substrate type, obstacle distribution, and boundary contour. Spatial data from several typical farming ponds were obtained through on-site surveys and corrected using remote sensing imagery to ensure the simulation environment closely approximates the real scene in scale and layout. Obstacles include fixed structures such as feeding piles and aerator supports, as well as dynamic disturbance sources simulating the effects of water flow drift or temporary floating objects.

[0028] This application provides simulation and experimental verification: The simulation platform integrates a physics engine to recreate the dynamic characteristics of the unmanned surface vessel (USV), including propulsion system response delay, rudder angle control accuracy, and the impact of wind and wave disturbances on course stability. The vessel motion model references the technical parameters of actual bait-dropping USVs, setting maximum speed, turning radius, and acceleration limits to make the simulated trajectory more closely resemble the actual operational behavior. The perception module simulates the onboard sensor system, including GPS positioning errors, ultrasonic obstacle avoidance detection range, and blind spots, enhancing the realism of environmental interaction.

[0029] The simulation software employs a collaborative architecture of Gazebo and ROS (Robot Operating System), leveraging its powerful multi-agent support and plug-in extensibility to embed and debug path planning algorithms in real time. Within this framework, the instructions output by the planning module directly drive the virtual unmanned vessel to perform navigation tasks, while simultaneously providing feedback on indicators such as position deviation, energy consumption, and task completion rate. To test adaptability under different operating conditions, several typical scenarios are preset: regular rectangular ponds, irregular multi-bay ponds, high-density obstacle areas, and sudden obstacle insertion scenarios.

[0030] A tidal variation model is introduced in the time dimension to simulate the impact of water level fluctuations on navigable areas. A noise interference model is set up for the communication link to test the robustness of the control system under unstable signal conditions. All scenarios are equipped with a visualization interface to facilitate observation of the path generation process, obstacle avoidance decision logic, and global exploration efficiency. Recorded data includes path length, number of turns, task time, fuel estimate, and deviation from the ideal path, for subsequent quantitative analysis. The entire simulation system has good reproducibility and parameter adjustment flexibility, providing a stable and reliable testing foundation for algorithm improvement.

[0031] The experimental design focuses on the practical application of the improved path planning algorithm in a mud crab farming pond environment, constructing a test system that closely resembles a real-world scenario. The experimental area simulates typical square and irregularly shaped farming ponds, with dimensions of 30 meters × 30 meters and a complex terrain featuring multiple sheltered areas and shallows. Static obstacles such as stone piles and buoys, as well as dynamic interference sources, are placed in the environment to simulate the effects of water flow drift and sudden floating objects. The unmanned surface vessel (USV) platform is equipped with differential GPS and ultrasonic sensor arrays, achieving centimeter-level positioning accuracy and multi-directional environmental perception capabilities. The communication module ensures real-time monitoring of the navigation status by the shore-based control unit.

[0032] The test task was set as a fixed-point, quantitative automatic feeding operation. Target points were distributed based on the high-frequency feeding activity areas of mud crabs, with a total of 12 feeding nodes forming a non-uniform spatial distribution pattern. The path planning system needed to complete a closed-loop task, starting from the initial point, traversing all feeding points, and returning while ensuring obstacle avoidance safety. Comparative experiments used three algorithms: the traditional A* algorithm, basic particle swarm optimization (PSO) combined with dynamic windowing (DWA), and the improved hybrid heuristic path planning method proposed in this paper. Each algorithm underwent 30 independent runs, recording key indicators such as complete path length, task execution time, energy consumption, obstacle avoidance success rate, and path smoothness.

[0033] The experiment was conducted in phases. The first phase verified the algorithm's basic performance under conditions without dynamic disturbance. The second phase introduced a controllable water flow simulation device to generate a directional flow field. The third phase introduced randomly moving obstacles to test the system's dynamic response capability. The data acquisition system simultaneously recorded trajectory coordinates, heading angle changes, speed adjustment curves, and decision response delays. To evaluate the algorithm's robustness, repeated tests were conducted at different wind speeds and water surface disturbance levels. Simultaneously, a visual interface was used to present the path generation process and sensor feedback information in real time, facilitating the tracing of anomalies.

[0034] The evaluation criteria comprehensively consider efficiency, safety, and practicality. Path length deviation rate is used to measure the optimality of the planning, time stability reflects the computational cost of the algorithm, and energy consumption is estimated through motor current integration. Safety indicators include minimum obstacle distance statistics and the number of collisions. Considering the characteristics of baiting operations, a functional evaluation item is added: the accuracy error range of the baiting point arrival (within ±0.5 meters is considered effective arrival). Experimental results will form a multidimensional dataset to support subsequent analysis of variance and significance testing, ensuring the statistical significance of the conclusions. The entire experimental procedure follows the principle of reproducibility, and parameter configurations and initial conditions are meticulously archived, providing a reference benchmark for similar studies.

[0035] The experiment constructs a two-dimensional static and dynamic obstacle hybrid scenario simulating a crab farming pond, and compares the proposed path planning algorithm with traditional A* and artificial potential field methods in multiple dimensions. The test environment includes obstacle layouts of varying densities, a variable number of target points, and the introduction of slowly moving disturbances to simulate water flow disturbances. Each of the three algorithms is run 30 times under the same initial conditions, recording four indicators: path length, obstacle avoidance success rate, task completion time, and path smoothness. The improved algorithm shortens the average path length by approximately 12.7% compared to A* and by 9.3% compared to the artificial potential field method. Its advantage stems from the adaptive weight strategy introduced in the global optimization stage, which dynamically adjusts the distance and safety term coefficients in the cost function, effectively avoiding the path detour phenomenon that easily occurs in traditional methods. The obstacle avoidance success rate reaches 98.6%, significantly higher than the 86.4% of the artificial potential field method, and no local minima are encountered. Regarding task completion time, the improved algorithm takes an average of 218 seconds, slightly longer than A*'s 203 seconds, but significantly better than the time consumed by the artificial potential field method due to repeated oscillations. Path curvature statistics show that the average rate of curvature change of the trajectory generated by the proposed method is reduced to 0.38 rad / m, indicating smoother control commands and improved stability of the unmanned surface vessel (USV). Further analysis reveals that in high obstacle density scenarios, while the traditional A* algorithm ensures integrity, the generated path has many right-angle turns, increasing turning energy consumption; the artificial potential field method exhibits multiple oscillations and adjustments when approaching the target area, affecting baiting accuracy. The improved algorithm, through a hierarchical architecture, achieves a balance between macroscopic path rationality and microscopic behavioral flexibility. In actual surface testing, the USV equipped with this algorithm stably completed the predetermined route over a five-day operational cycle, with GPS trajectory data deviation from the planned path controlled within ±0.4 meters. Overall, the proposed method improves overall operational efficiency while ensuring safety, meeting the needs of frequent, precise, and low-disturbance baiting in the crab farming environment. The algorithm demonstrates good robustness and practicality, providing a feasible technical solution for small USVs performing periodic tasks in confined waters.

[0036] A path planning device for an unmanned surface vessel, comprising: The initial adjustment module is used to collect environmental parameters within the pond area, construct a global map, input the geographical coordinates of the task area, set the distribution of feeding points, and adjust navigation parameters. The global route planning module calculates the route based on an algorithm that integrates a dynamic weighting mechanism and an adaptive cost function, generating a global route plan. The local obstacle avoidance module is used to replan local trajectories; The feeding module is used to periodically upload the location status and health indicators of the unmanned vessel; when it reaches the designated feeding point, it activates the feeding device, releases feed according to the preset dosage, and records the feeding time and coordinate information for later traceability.

[0037] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

Claims

1. A path planning method for an unmanned surface vessel, characterized in that: Comprising the following steps: Step 1: Collect the environmental parameters of the in-pond area, build a global map, and record the geographic coordinates of the task area; Set the bait distribution point and adjust the navigation parameters; Step 2: Obtain the current position, define the safe navigation boundary, and ensure that the subsequent movement does not exceed the preset range; Step 3: Based on the algorithm of fusion dynamic weight mechanism and adaptive cost function, solve the route and generate global route planning; Step 4: Decompose the global path into several navigation segments and calculate the target heading angle and propulsion speed of the current navigation segment in real time; Step 5: Continuously collect environmental data and input the data into the local obstacle avoidance module. If a dynamic obstacle is detected, the local trajectory is re-planned, and the global path is incrementally updated; Step 6: Periodically upload the unmanned ship position state and equipment health indicators. When reaching the specified bait distribution point, activate the feeding device, release the feed according to the preset dose, and record the release time and coordinate information for later tracing; Step 7: After completing bait distribution at the current bait distribution point, determine whether there are still unfinished tasks. If yes, proceed to the next bait distribution point. If no, start the return program and return to the starting position along the shortest safe path.

2. The path planning method of claim 1, wherein: The global path planning method is Step 31: Build a hybrid hierarchical model and pre-generate a rough guide path; Step 32: Based on environmental uncertainty factors, dynamically adjust the rough guide path to generate global route planning; Step 33: Use genetic algorithm to determine the access sequence of bait distribution points.

3. The path planning method of claim 2, wherein: The method of step 31: Based on the hybrid modeling method of grid map and topological map, a hybrid hierarchical model is constructed. The bottom layer of the model uses a grid map to describe the pond terrain and obstacle location; The model extracts key landmark points of the navigable area to form a simplified topological graph. The IA* algorithm runs on the bottom grid, and the Dijkstra algorithm is used on the top layer to pre-generate a rough guide path.

4. The path planning method of claim 2, wherein: The method of step 32 is: Let the current node be "n", and the total cost function be defined as f(n) = g(n) + \omega(n) \cdot h(n); Where "g(n)" represents the actual path cost from the starting point to node "n"; "h(n)" is the heuristic value estimated by Euclidean or Manhattan distance; "ω(n)" is a dynamic weight coefficient that changes with the environment state, "ω(n)" is adjusted online according to the local obstacle density "ρ(n)" and the water flow velocity vector "v_w(n)", the formula is ω(n) = 1 + \alpha \cdot \frac{\rho(n)}{\rho_{max}} + \beta \cdot \|\vec{v}_w(n)\| Where "α" and "β" are empirical tuning coefficients, controlling the sensitivity of obstacle density and water flow, respectively, with a value range of [0.5, 1.2]. When the obstacle is dense or the water flow is strong in a certain area, the heuristic weight of the nodes in that area is automatically increased.

5. The path planning method of claim 2, wherein: The method of step 33 is: The path planning is decomposed into two stages: sequence optimization and single segment path generation; The first stage sequence optimization: a traveling salesman problem with time windows (TSP-TW) is constructed, and a genetic algorithm is used to solve the visiting sequence with the minimum total distance: The chromosome coding adopts an integer permutation form, and each gene represents a feeding point number. The fitness function is defined as: F(\pi) = \frac{1}{C(\pi) + \gamma \cdot T_{violation}} Where " C(\pi) " is the total path length, " T_{violation} " is the cumulative length of time window constraint violation, and " \gamma " is the penalty coefficient. The crossover operation adopts order crossover (OX), and the mutation adopts reverse order mutation to maintain the validity of the solution. After multiple generations of evolution, the optimal visiting sequence " \pi^* " is output as the input for subsequent single segment path planning; The second stage single segment path generation: based on the sequence, IA* planning is performed, and path smoothing processing is introduced. A method combining cubic spline interpolation and Bezier curve fitting is used for post-processing. For any three consecutive path points "P_{i-1}, P_i, P_{i+1} ", a quadratic Bezier curve is constructed: B(t) = (1-t)^2 P_{i-1} + 2t(1-t)P_c + t^2 P_{i+1}, \quad t \in [0,1] Where the control point " P_c " is set to " (P_{i-1} + P_{i+1}) / 2 ", ensuring curvature continuity. At the same time, it is checked whether each curve segment crosses the obstacle buffer zone. If there is a conflict, an intermediate waypoint is inserted for re-fitting.

6. The path planning method of claim 1, wherein: The local obstacle avoidance module of step 5 adopts a dynamic window algorithm, and the dynamic obstacle is a sudden obstacle. When the dynamic obstacle causes the current unmanned ship path to be interrupted, the local trajectory is re-planned.

7. The path planning method of claim 1, wherein: The method of step 5 is: When the unmanned ship approaches an obstacle on the pre-planned path and the distance is less than the threshold " d_{safe} = 2m ", the local re-planning mechanism is triggered; DWA samples several feasible motion commands in the velocity space " (v, \omega) ". The optimal instantaneous action is selected according to the evaluation function: U(v, \omega) = w_1 \cdot \text{heading}(v, \omega) + w_2 \cdot \text{dist}(v, \omega) + w_3 \cdot \text{vel}(v, \omega) Where the heading term measures the angular deviation of the forward direction towards the target, the dist term reflects the distance to the nearest obstacle, and the vel term encourages maintaining a high linear velocity. The weights " w_i " are dynamically adjusted according to the task stage: in open water, increase " w_3 " to increase the speed, and in narrow channels, increase " w_2 " to ensure safety. 8.A path planning device of an unmanned ship, characterized by comprising: Comprise An initial adjustment module is configured to collect environmental parameters of the area in the pond, build a global map, and record geographic coordinates of the task area; set bait distribution and navigation parameter adjustment; The global route planning module solves the route based on an algorithm of a dynamic weight fusion mechanism and a self-adaptive cost function, and generates a global route plan; A local obstacle avoidance module is configured to re-plan a local trajectory; A baiting module is configured to periodically upload the position state of the unmanned ship and the equipment health index; when reaching a specified baiting point, the baiting device is activated, the feed is released according to a preset dose, and the baiting time and coordinate information are recorded for later tracing.

9. A path planning system for an unmanned ship, characterized by: The computer program is stored on a memory and executed by a processor, and when the computer program is executed by the processor, the computer program performs the path planning method of the unmanned ship according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the path planning method of the unmanned ship according to any one of claims 1-7.

Images