A DQN-based optimal path optimization method for harvesting seaweed based on proximity target guidance

Abstract

The present application relates to the technical field of marine environment management, and particularly relates to a DQN optimal path optimization method for green algae salvage based on adjacent target guidance, comprising the following steps: S1, fusing green algae drift trajectory prediction and ship dynamics response, constructing a time-space coupled salvage weight correction matrix, and dynamically reflecting the matching degree of drift displacement and turning ability; S2, selecting the current optimal unvisited unit as an adjacent target based on the correction matrix, predicting the green algae position when the ship arrives, and generating a guidance action set limited by the minimum turning radius; S3, constructing a double-channel DQN network, the main channel extracts task accessibility, the auxiliary channel depicts the drift change trend, and the optimal salvage path suitable for complex drift environment is generated through a dynamic attenuation fusion mechanism. Through the construction of the green algae-ship coupling model and the acceleration perception guidance mechanism, the present application realizes the collaborative improvement of the feasibility, stability and environmental adaptability of path planning.

Description

Technical Field

[0001] This invention relates to the field of marine environmental management technology, and in particular to an optimal path optimization method for DQN seaweed harvesting based on proximity target guidance. Background Technology

[0002] With the increasingly severe global marine ecological environment problems, the frequent occurrence of seaweed disasters has become an important factor affecting nearshore fishery resources, navigation safety and the stability of coastal ecosystems. In order to cope with the environmental pollution and shipping interference caused by large-scale seaweed accumulation, researching intelligent and efficient seaweed harvesting path optimization methods has become an important direction for current marine environmental governance. Affected by complex factors such as ocean currents and wind fields, seaweed has significant drift characteristics. Traditional harvesting strategies based on static area priority or fixed paths are no longer able to meet the requirements of real-time and adaptability.

[0003] Existing methods for optimizing the harvesting path of seaweed mainly rely on historical drift trends or static target segmentation, failing to effectively consider the coupling relationship between the uncertainty of seaweed drift and ship dynamic constraints. This can easily lead to problems such as path deviation, turning failure, or decreased harvesting efficiency. In addition, although some path planning models based on reinforcement learning have a certain degree of autonomy, they do not introduce a nearby target guidance mechanism and turning constraint, making it difficult to generate highly feasible harvesting paths in real time in dynamic environments. Furthermore, they lack an adaptive control mechanism for the intensity of seaweed drift, affecting the model's generalization ability and decision stability in rapid drift situations. Summary of the Invention

[0004] This invention provides a DQN-based optimal path optimization method for seaweed harvesting based on proximity target guidance. It integrates seaweed drift trajectory prediction with a spatiotemporally coupled harvesting weight correction mechanism constructed from a ship response model, generates guiding actions by combining minimum turning radius constraints, and achieves adaptive path optimization for rapidly changing environments by introducing a dual-channel DQN structure and a dynamic attenuation fusion strategy for drift perception. This improves the feasibility, stability, and environmental responsiveness of the harvesting path.

[0005] A method for optimizing the optimal path for *Ulva prolifera* harvesting based on proximity target guidance (DQN) includes the following steps:

[0006] S1, Input the real-time drift vector field of seaweed and ship dynamics parameters, and generate a spatiotemporally coupled salvage weight correction matrix by fusing seaweed drift trajectory prediction with ship motion response characteristics. The weight value of each unit of the salvage weight correction matrix is ​​dynamically adjusted according to the matching degree between the drift displacement of seaweed during the ship's navigation time and the ship's turning ability.

[0007] S2, based on the salvage weight correction matrix, select the unvisited salvage unit with the highest current weight as the nearest target, and predict the actual location of the seaweed when the ship arrives at the nearest target, and generate a set of guiding actions with ship turning curvature constraints. The direction range of the guiding actions is dynamically limited according to the minimum turning radius of the ship to eliminate the path deviation caused by turning inertia.

[0008] S3. Construct a dual-channel DQN network. The main channel inputs the current ship status and the salvage weight correction matrix, while the auxiliary channel inputs the time change rate of the seaweed drift trend. The dual-channel outputs are balanced through a dynamic attenuation fusion mechanism to generate the final salvage path.

[0009] Optionally, S1 includes:

[0010] S11 acquires real-time data on the drift vector field of seaweed and the dynamic parameters of the ship itself, including speed, acceleration, minimum turning radius and response delay. Based on the trajectory prediction model, it predicts the drift path of seaweed in the future period. At the same time, it introduces the ship dynamic response model to describe the evolution process of its motion trajectory under given input conditions and constructs a seaweed-ship motion coupling expression model including time and space dimensions.

[0011] S12. Using the constructed seaweed-ship motion coupling expression model, all grid cells in the salvage area are traversed one by one. Based on the dynamic coverage trajectory of the ship during the predetermined navigation time, the actual coverage efficiency of each target cell is calculated. Combining the degree of matching between the drift displacement of seaweed during the predetermined navigation time and the ship's turning ability, the initial salvage weight is dynamically corrected, and a salvage weight correction matrix reflecting the salvage feasibility and real-time adaptability of each cell is constructed.

[0012] Optionally, S11 includes:

[0013] S111 uses remote sensing images, drifting buoys, and shipboard navigation systems to collect real-time data on the drift vector field of seaweed and ship dynamics parameters, including speed, acceleration, minimum turning radius, and response delay.

[0014] S112, based on the collected drift vector field, uses the time-series integration method to dynamically predict the future drift path of Ulva prolifera and outputs a sequence of Ulva prolifera trajectory points;

[0015] S113. By combining the ship's dynamic response model, the future response trajectory is constructed, and the relative distance between the seaweed and the ship is calculated. Finally, a coupled expression model of seaweed-ship motion, including time and space dimensions, is constructed. .

[0016] Optionally, S12 includes:

[0017] S121, Based on the Coupled Expression Model of Ulva prolifera-ship motion The salvage area is divided into a set of discrete grid cells. According to each unit The corresponding drift trajectory segment of the seaweed With ship response trajectory segment The timing correspondence, extraction unit Relative position information at all time steps;

[0018] S122, for each unit The ship's response position at time step i The location of the centroid of *Ulva prolifera* in this unit Euclidean distance As a criterion, the calculation unit Time-weighted coverage efficiency ;

[0019] S123. Based on the directional change of the drift trajectory of seaweed in each grid cell, calculate its turning fluctuation index, and construct a matching factor in combination with the minimum turning radius of the ship to measure the difficulty of the ship following the trajectory. Finally, the matching factor and the coverage efficiency are applied together to the initial salvage weight to complete the dynamic correction of the weight of each cell and generate a salvage weight correction matrix that reflects the path adaptability.

[0020] Optionally, S2 includes:

[0021] S21. Based on the constructed salvage weight correction matrix, the cell with the largest current weight value is selected from all unvisited grid cells as the nearest target, representing the candidate area with the highest salvage priority at the current moment. Based on the sailing time between the ship's current position and the target cell, combined with the predicted drift trajectory of the seaweed, the actual position of the centroid of the seaweed when the ship is expected to arrive at the cell is calculated.

[0022] S22. Based on the predicted target location of the seaweed, combined with the ship's current position and orientation, a sequence of guiding actions is generated to guide the ship toward the target. At the same time, the minimum turning radius constraint of the ship is considered to limit the range of the guiding direction.

[0023] Optionally, S21 includes:

[0024] S211, define the set of all grid cells within the salvage area as... Let the set of currently visited cell indices be... Then the set of unvisited units is Extract the corresponding set of corrected weights. ;

[0025] S212, Select the cell index corresponding to the maximum value from the unvisited modified weight set. As the current nearest target, the corresponding nearest target grid cell is Its current center position coordinates are ;

[0026] S213, Let the current ship position be... The ship's speed is The estimated time to reach the target unit is then... Based on the trajectory prediction function of seaweed drift Calculated in The position of the center of mass of the moss at a given moment.

[0027] Optionally, S22 includes:

[0028] S221, Let the current position of the ship be... Orientation angle is The predicted actual location of the seaweed is Then, based on the ship's heading angle to the target Calculate the relative angle difference between the current orientation and the target orientation. ;

[0029] S222, based on the minimum turning radius of the ship Calculate the maximum yaw angle constrained by steering constraints within a given time window. And limit the range of guide angles. ;

[0030] S223, with discrete intervals based on guide angle The current ship orientation near the set of steering angles Each steering angle is mapped to a corresponding navigation step, and the target position after the ship has traveled a certain distance in that direction is calculated, ultimately forming a steering action sequence that includes the steering angle and the target coordinates. .

[0031] Optionally, S3 includes:

[0032] S31. Establish a dual-channel DQN network. The main channel inputs the current ship status and salvage weight correction matrix to extract reachability and salvage priority information. The auxiliary channel inputs the time change rate of the seaweed drift trend to identify the instability and change trend of the drift.

[0033] S32 uses a dynamic attenuation mechanism driven by drift acceleration to balance the Q values ​​of the main and auxiliary channels. When the drift acceleration of seaweed exceeds a set threshold, the weight of the auxiliary channel in the fusion is increased, and the optimal retrieval path is finally output.

[0034] Optionally, S31 includes:

[0035] S311, the main channel input consists of the current vessel's state vector and the salvage weight correction matrix. The state vector includes position, orientation and speed information. The salvage weight correction matrix is ​​flattened into a vector and then concatenated with the state vector to serve as the feature vector of the main channel.

[0036] S312, the auxiliary channel extracts the rate of change characteristics of the drift trend in the region by calculating the velocity and acceleration of the drift trajectory of each grid cell of Ulva lactuca, and calculates its average drift acceleration in the x and y directions as the feature vector of the auxiliary channel;

[0037] S313, the feature vectors of the main channel and the feature vectors of the auxiliary channel are fed into independent DQN branch networks, and the Q-value vectors corresponding to the guide action sets are output respectively.

[0038] Optionally, S32 includes:

[0039] S321, based on the average acceleration vector of the entire region's *Ulva prolifera*, calculates the total acceleration modulus, which serves as an indicator of the drift intensity in the current environment. ;

[0040] S322, Set control threshold ,when At the same time, the influence of the auxiliary channel is enhanced, and an exponential decay function is constructed based on the drift intensity index and the control threshold to dynamically adjust the fusion weight coefficient of the main and auxiliary channels;

[0041] S323 integrates the dual-channel Q values ​​and selects the guiding action corresponding to the maximum value to output the current optimal salvage path decision.

[0042] The beneficial effects of this invention are:

[0043] This invention constructs a salvage weight correction matrix that integrates the prediction of seaweed drift trajectory and ship dynamics constraints. This matrix can dynamically reflect the coupling relationship between the seaweed movement trend and ship accessibility, effectively overcoming the problem of misalignment between path generation and actual salvage target in traditional methods, and improving the feasibility of salvage path execution and prediction accuracy.

[0044] This invention, by introducing a guidance action generation mechanism that combines proximity target screening and minimum turning radius constraints, can avoid target deviation and turning failure caused by the ship's turning inertia, ensure that each path is physically feasible under the current attitude, and enhance the stability and local path adaptation capability during the salvage process.

[0045] This invention constructs a dual-channel DQN structure driven by the rate of change of drift trend and introduces a dynamic decay fusion mechanism based on acceleration sensing. This enhances the influence of the sensing channel on strategy generation when seaweed drifts drastically, significantly improving the robustness and environmental adaptability of salvage path planning under complex sea conditions. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the optimization method flow according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the salvage weight correction process according to an embodiment of the present invention. Detailed Implementation

[0049] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0050] like Figures 1-2 As shown, a DQN-based optimal path optimization method for harvesting seaweed based on proximity target guidance includes the following steps:

[0051] S1, input the real-time drift vector field of seaweed and ship dynamics parameters, and generate a spatiotemporally coupled salvage weight correction matrix by integrating the seaweed drift trajectory prediction and ship motion response characteristics. The weight value of each unit of the salvage weight correction matrix is ​​dynamically adjusted according to the matching degree between the drift displacement of seaweed during the ship's navigation time and the ship's turning ability.

[0052] S2, based on the salvage weight correction matrix, selects the unvisited salvage unit with the highest current weight as the nearest target, and predicts the actual location of the seaweed when the ship arrives at the nearest target, generating a set of guiding actions with ship turning curvature constraints. The direction range of the guiding actions is dynamically limited according to the minimum turning radius of the ship to eliminate path deviation caused by turning inertia.

[0053] S3. Construct a dual-channel DQN network. The main channel inputs the current ship status and the salvage weight correction matrix, while the auxiliary channel inputs the time change rate of the seaweed drift trend. The outputs of the two channels are balanced through a dynamic attenuation fusion mechanism to generate the final salvage path. The decision weight of the auxiliary channel increases with the increase of the seaweed drift acceleration to ensure that the path dynamically adapts to the rapidly changing environment.

[0054] S1 includes:

[0055] S11 acquires real-time data on the drift vector field of seaweed and the dynamic parameters of the ship itself, including speed, acceleration, minimum turning radius and response delay. Based on the trajectory prediction model, it predicts the drift path of seaweed in the future period. At the same time, it introduces the ship dynamic response model to describe the evolution process of its motion trajectory under given input conditions and constructs a seaweed-ship motion coupling expression model including time and space dimensions.

[0056] S12. Using the constructed seaweed-ship motion coupling expression model, all grid cells in the salvage area are traversed one by one. Based on the dynamic coverage trajectory of the ship during the predetermined navigation time, the actual coverage efficiency of each target cell is calculated. Combining the degree of matching between the drift displacement of seaweed during the predetermined navigation time and the ship's turning ability, the initial salvage weight is dynamically corrected, and a salvage weight correction matrix reflecting the salvage feasibility and real-time adaptability of each cell is constructed.

[0057] S11 includes:

[0058] S111, through remote sensing imagery, drifting buoys, and shipborne navigation systems, collects real-time data on the drift vector field of *Ulva prolifera* and ship dynamics parameters, including speed, acceleration, minimum turning radius, and response delay. Specifically, this includes:

[0059] Ulva prolifera drift vector field data: derived from satellite remote sensing image sequences, UAV observations, or drifting buoys on the sea surface, a two-dimensional vector field is obtained through image recognition and physical field inversion methods, represented as:

[0060] ;

[0061] in, For the location of seaweed The drift velocity vector at time t, , These are the drift velocity components in the x-axis and y-axis directions, respectively;

[0062] Ship dynamic parameters: acquired in real time by the ship's onboard inertial measurement unit (IMU), GPS, and ship control system, including current speed. acceleration Minimum turning radius Response delay ;

[0063] S112, based on the collected drift vector field, uses the time-series integration method to dynamically predict the future drift path of *Ulva prolifera*, outputting a sequence of *Ulva prolifera* trajectory points, represented as:

[0064] ;

[0065] in, Let be the center position of the seaweed at the current time t. To predict the time step;

[0066] ;

[0067] in, To predict the number of steps, satisfying , The expected sailing time of the vessel. This is a sequence of trajectory points for *Ulva prolifera*.

[0068] S113. By combining the ship's dynamic response model, the future response trajectory is constructed, and the relative distance between the seaweed and the ship is calculated. Finally, a coupled expression model of seaweed-ship motion, including time and space dimensions, is constructed. , is represented as:

[0069] ;

[0070] ;

[0071] ;

[0072] in, For the ship at the i-th predicted time in the future The position coordinates along the x-axis and y-axis. For the ship's dynamic response trajectory function, Let i be the time increment of the i-th prediction time relative to the current time. For the sequence of ship response trajectory points, Let the coordinates be the center position of the seaweed at the i-th prediction time step. Let be the coordinates of the ship's response position at the i-th prediction time step. Let be the Euclidean relative distance between the seaweed and the ship at the i-th time point;

[0073] Ship dynamic response trajectory function Represented as:

[0074] ;

[0075] in, , This represents the ship's current position (starting point). Let be the ship's current heading angle, and k be the discrete iteration step number of the cumulative prediction time step.

[0076] S12 includes:

[0077] S121, Based on the Coupled Expression Model of Ulva prolifera-ship motion The salvage area is divided into a set of discrete grid cells. According to each unit The corresponding drift trajectory segment of the seaweed With ship response trajectory segment The timing correspondence, extraction unit Relative position information at all time steps;

[0078] S122, for each unit The ship's response position at time step i The location of the centroid of *Ulva prolifera* in this unit Euclidean distance As a criterion, the calculation unit Time-weighted coverage efficiency , is represented as:

[0079] ;

[0080] ;

[0081] in, Let i be the coverage indicator function at time i. The radius of the salvage operation;

[0082] S123, based on the directional changes of the *Ulva prolifera* drift trajectory within each grid cell, calculate its turning fluctuation index, and construct a matching factor in conjunction with the ship's minimum turning radius to measure the ease with which the ship follows the trajectory. Finally, the matching factor and coverage efficiency are applied together to the initial salvage weights to dynamically correct the weights of each cell, generating a salvage weight correction matrix that reflects path adaptability, expressed as:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] in, Let i be the turning angle of the drift direction of the seaweed at time step i. Let be the drift direction vector of the seaweed in the j-th unit at the i-th step. Let be the drift direction vector of the seaweed in the (i-1)th step of the j-th unit. , These represent the predicted positions of the seaweed in unit j at steps i and i-1, respectively. For unit The average turning fluctuation angle, This is the steering sensitivity adjustment coefficient. For unit Steering matching factor As the initial salvage weight, To correct the salvage weights, To retrieve the weight correction matrix, This represents the total number of grid cells after the salvage area has been divided.

[0090] S2 includes:

[0091] S21. Based on the constructed salvage weight correction matrix, the cell with the largest current weight value is selected from all unvisited grid cells as the nearest target, representing the candidate area with the highest salvage priority at the current moment. Based on the sailing time between the ship's current position and the target cell, combined with the predicted drift trajectory of the seaweed, the actual position of the centroid of the seaweed when the ship is expected to arrive at the cell is calculated.

[0092] S22. Based on the predicted target location of the seaweed, combined with the ship's current position and orientation, a sequence of guiding actions is generated to guide the ship toward the target. At the same time, the minimum turning radius constraint of the ship is considered to limit the range of the guiding direction, thereby avoiding path deviation or turning failure caused by turning inertia, and ensuring the feasibility and stability of path execution.

[0093] S21 includes:

[0094] S211, define the set of all grid cells within the salvage area as... Let the set of currently visited cell indices be... Then the set of unvisited units is Extract the corresponding set of corrected weights. , is represented as:

[0095] ;

[0096] in, The salvage adjustment weight for the j-th unit;

[0097] S212, Select the cell index corresponding to the maximum value from the unvisited modified weight set. As the current nearest target, the corresponding nearest target grid cell is Its current center position coordinates are ;

[0098] S213, Let the current ship position be... The ship's speed is The estimated time to reach the target unit is then... Based on the trajectory prediction function of seaweed drift Calculated in The position of the centroid of the moss at a given time is represented as:

[0099] ;

[0100] ;

[0101] in, For the currently selected neighboring target unit In the example, the drift prediction trajectory function of the centroid position of *Ulva prolifera* over time;

[0102] Ulva prolifera drift prediction trajectory function Represented as:

[0103] ;

[0104] in, For the j-th unit, in the k-th prediction step in the future (time is...) Predicted location of the centroid of *Ulva prolifera* at that time. Let m be the time step and m be the summation variable. In the j-th unit, at time The drift velocity vector of the seaweed at that location.

[0105] S22 includes:

[0106] S221, Let the current position of the ship be... Orientation angle is The predicted actual location of the seaweed is Then, based on the ship's heading angle to the target Calculate the relative angle difference between the current orientation and the target orientation. , is represented as:

[0107] ;

[0108] ;

[0109] S222, based on the minimum turning radius of the ship Calculate the maximum yaw angle constrained by steering constraints within a given time window. And limit the range of guide angles. , is represented as:

[0110] ;

[0111] Where L is the travel distance corresponding to the step size generated by the guiding action;

[0112] S223, with discrete intervals based on guide angle The current ship orientation near the set of steering angles Each steering angle is mapped to a corresponding navigation step, and the target position after the ship has traveled a certain distance in that direction is calculated, ultimately forming a steering action sequence that includes the steering angle and the target coordinates. , is represented as:

[0113] ;

[0114] ;

[0115] ;

[0116] in, The target angle for discrete guiding actions. The next position after performing the guiding action.

[0117] S3 includes:

[0118] S31. Establish a dual-channel DQN network. The main channel inputs the current ship status and salvage weight correction matrix to extract reachability and salvage priority information. The auxiliary channel inputs the time change rate of the seaweed drift trend to identify the instability and change trend of the drift.

[0119] S32 uses a dynamic attenuation mechanism driven by drift acceleration to balance the Q values ​​of the main and auxiliary channels. When the drift acceleration of seaweed exceeds a set threshold, the weight of the auxiliary channel in the fusion is increased to ensure that the path strategy can dynamically adapt to environmental changes and ultimately output the optimal salvage path.

[0120] S31 includes:

[0121] S311, the main channel input consists of the current vessel's state vector and the salvage weight correction matrix. The state vector includes position, orientation, and velocity information. The salvage weight correction matrix, flattened into a vector, is concatenated with the state vector to serve as the main channel's feature vector. This vector is used to extract the current environmental reachability and task priority, expressed as:

[0122] ;

[0123] in, , The ship's current position. The ship's current orientation. Current speed;

[0124] ;

[0125] in, The feature vector of the main channel. This is the flattened weight vector;

[0126] S312, the auxiliary channel extracts the rate of change characteristics of the drift trend within the region by calculating the velocity and acceleration of the drift trajectory of *Ulva prolifera* in each grid cell, and calculates its average drift acceleration in the x and y directions as the feature vector of the auxiliary channel, expressed as:

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] ;

[0132] in, Let Vulva be the drift velocity vector. , These are the centroid positions of the *Ulva prolifera* at the (i+1)th and i-th steps of the j-th unit, respectively. The feature vector of the auxiliary channel. , The components of the average drift acceleration of all units, , These are the drift acceleration components of the j-th unit in the x and y directions, respectively;

[0133] S313, the feature vectors of the main channel and the auxiliary channel are fed into independent DQN branch networks, respectively, and the Q-value vectors corresponding to the guidance action sets are output, as follows:

[0134] ;

[0135] ;

[0136] in, , These represent the action Q values ​​of the main channel and auxiliary channel outputs, respectively.

[0137] S32 includes:

[0138] S321, based on the average acceleration vector of the entire region's *Ulva prolifera*, calculates the total acceleration modulus, which serves as an indicator of the drift intensity in the current environment. , is represented as:

[0139] ;

[0140] in, This is an indicator of drift strength.

[0141] S322, Set control threshold ,when At the same time, the influence of the auxiliary channel is enhanced, and an exponential decay function is constructed based on the drift intensity index and the control threshold to dynamically adjust the fusion weight coefficient of the main and auxiliary channels, expressed as:

[0142] ;

[0143] ;

[0144] in, This is the attenuation adjustment coefficient. Main channel weighting factor, As a secondary channel weighting factor;

[0145] ;

[0146] in, This is the sensitivity adjustment coefficient. This represents the typical cruising speed of the ship.

[0147] S323, by fusing the dual-channel Q values ​​and selecting the guiding action corresponding to the maximum value, outputs the current optimal salvage path decision, expressed as:

[0148] ;

[0149] ;

[0150] in, The fused Q-value sequence, For the optimal guiding action, K is the total number of actions in the current set of candidate guiding actions.

[0151] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0152] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A DQN optimal path optimization method for green algae salvage based on target proximity guidance, characterized in that, The method comprises the following steps: S1, inputting real-time Enteromorpha drift vector field and ship dynamics parameters, and generating a time-space coupled salvage weight correction matrix by fusing Enteromorpha drift trajectory prediction and ship motion response characteristics, wherein each unit weight value of the salvage weight correction matrix is dynamically adjusted according to the matching degree between the drift displacement of Enteromorpha within the ship sailing time and the turning ability of the ship; S2, selecting the currently highest-weight unvisited salvage unit as the adjacent target based on the salvage weight correction matrix, and predicting the actual Enteromorpha position when the ship reaches the adjacent target, and generating a set of guidance actions with ship turning curvature constraints, wherein the direction range of the guidance actions is dynamically limited according to the minimum turning radius of the ship to eliminate the path deviation caused by turning inertia; S3, constructing a double-channel DQN network, inputting the current ship state and the salvage weight correction matrix into the main channel, inputting the time variation rate of the Enteromorpha drift trend into the auxiliary channel, balancing the double-channel output through a dynamic attenuation fusion mechanism, and generating the final salvage path; The S1 comprises: S11, acquiring real-time Enteromorpha drift vector field data and ship's own dynamics parameters, including speed, acceleration, minimum turning radius and response delay, predicting the drift path of Enteromorpha within the future period based on a trajectory prediction model, introducing a ship dynamics response model to describe the motion trajectory evolution process under given input conditions, and constructing an Enteromorpha-ship motion coupling expression model including time and space dimensions; S12, using the constructed Enteromorpha-ship motion coupling expression model to traverse all grid units in the salvage area one by one, calculating the actual coverage efficiency of each target unit according to the dynamic coverage trajectory of the ship within the predetermined sailing time, combining the matching degree between the drift displacement of Enteromorpha within the predetermined sailing time and the turning ability of the ship, dynamically correcting the initial salvage weight, and constructing a salvage weight correction matrix reflecting the salvage feasibility and real-time adaptability of each unit; The S12 comprises: S121, Based on the Coupled Expression Model of Ulva prolifera-ship motion The salvage area is divided into a set of discrete grid cells. According to each unit The corresponding drift trajectory segment of the seaweed With ship response trajectory segment The timing correspondence, extraction unit Relative position information at all time steps; S122, for each unit The ship's response position at time step i The location of the centroid of *Ulva prolifera* in this unit Euclidean distance As a criterion, the calculation unit Time-weighted coverage efficiency ; S123, calculating the turning fluctuation index according to the direction change of the Enteromorpha drift trajectory in each grid unit, combining the minimum turning radius of the ship to construct a matching factor for measuring the difficulty of the ship following the trajectory, and finally applying the matching factor and the coverage efficiency to the initial salvage weight to complete the dynamic correction of the weight of each unit, and generate a salvage weight correction matrix reflecting the path adaptability; The S2 comprises: S21, selecting the unit with the largest weight value from all unvisited grid units as the adjacent target based on the constructed salvage weight correction matrix, representing the candidate area with the highest salvage priority at the current time, and calculating the actual position of the Enteromorpha centroid when the ship is expected to reach the unit according to the sailing time between the current position of the ship and the target unit and the drift prediction trajectory of Enteromorpha; S22, generating a guidance action sequence for guiding the ship towards the target according to the predicted target Enteromorpha position, combining the current position and direction of the ship, and considering the minimum turning radius constraint of the ship to limit the range of the guidance direction.

2. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 1, characterized in that, The S11 comprises: S111, real-time collection of drift vector field data and ship dynamics parameters including speed, acceleration, minimum turning radius and response delay through remote sensing images, drift buoys and ship navigation systems; S112, based on the collected drift vector field, the future drift path of Enteromorpha is dynamically predicted by using time series integration method, and a sequence of trajectory points of Enteromorpha is outputted; S113, combined with the ship dynamics response model to construct the future response trajectory, and calculate the relative distance between Enteromorpha and the ship, finally, construct the Enteromorpha-ship motion coupling expression model including time and space dimensions .

3. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 2, characterized in that, The S21 comprises: S211, define all grid cell sets in the fishing area as , set the current visited cell index set as , the unvisited cell set is , extract the corresponding correction weight set ; S212, selecting the cell index corresponding to the maximum value from the unvisited correction weight set As the current neighboring target, the corresponding neighboring target grid cell is , and the current center position coordinates of which are ; S213, if the current ship position is , and the ship speed is , the estimated time to reach the target cell is , the position of the green tide centroid at is calculated according to the green tide drift prediction trajectory function .

4. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 3, characterized in that, The S22 comprises: S221, assuming the current position of the ship is , the heading angle is , and the actual position of the Enteromorpha predicted is , then the relative angle difference between the current heading and the target direction is calculated according to the direction angle of the ship to the target . ; S222, calculating a maximum deflectable angle under the constraints of the minimum turning radius of the vessel for a given time window and defining a range of steering angles ; S223, at the heading angle discrete intervals In the current ship heading towards the nearby structure to guide the angle set And each guide angle is mapped to the corresponding one-step navigation action, the target position of the ship after advancing a distance in this direction is calculated, and finally a guide action sequence including the guide angle and the target coordinate is formed .

5. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 4, characterized in that, The S3 comprises: S31, a double-channel DQN network is established, the main channel inputs the current ship state and the salvage weight correction matrix, extracts the accessibility and salvage priority information, and the auxiliary channel inputs the time variation rate of Enteromorpha drift trend, which is used to identify the instability and change trend of drift; S32, a dynamic attenuation mechanism driven by drift acceleration is used to balance the Q values outputted by the main and auxiliary channels, when the drift acceleration of Enteromorpha exceeds the set threshold, the weight of the auxiliary channel in fusion is increased, and finally the optimal salvage path is outputted.

6. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 5, characterized in that, The S31 comprises: S311, the main channel inputs the state vector of the current ship and the salvage weight correction matrix, wherein the state vector includes position, orientation and speed information, and the salvage weight correction matrix is flattened into a vector and spliced with the state vector as the feature vector of the main channel; S312, the auxiliary channel extracts the change rate features of the drift trend in the region by calculating the speed and acceleration of the Enteromorpha drift trajectory of each grid element, and calculates the average drift acceleration in the x and y directions as the feature vector of the auxiliary channel; S313, the feature vectors of the main and auxiliary channels are respectively sent into independent DQN branch networks, and the Q value vectors corresponding to the sets of guiding actions are respectively outputted.

7. The DQN optimal path optimization method for harvesting Enteromorpha based on target proximity guidance according to claim 6, characterized in that, The S32 comprises: S321, based on the average vector of the total region enteromorpha acceleration, calculate the total acceleration module length as the drift intensity index of the current environment ; S322, set the control threshold When , the influence of the enhanced secondary channel is enhanced, and an exponential decay function is constructed according to the drift intensity index and the control threshold to dynamically adjust the fusion weight coefficient of the primary and secondary channels. S323, the double-channel Q values are fused and the guiding action corresponding to the maximum value is selected, and the current optimal salvage path decision is outputted.

Images