Intelligent prediction method for ship navigation trajectory

Abstract

The invention discloses a ship navigation trajectory intelligent prediction method, which belongs to the technical field of ship navigation, and comprises the following steps: dynamically acquiring and fusing multi-source navigation data such as a real-time state of a ship, a historical navigation mode, current and forecast environmental factors and the like; constructing and updating a dynamic feature vector representing the space-time motion behavior and intention of the ship; inputting the vector into a pre-trained integrated deep learning model to generate a probabilistic trajectory prediction sequence; according to the sequence, potential collision and yaw risks are assessed in combination with an electronic chart and traffic rules, and a comprehensive risk assessment index is output; and when the risk exceeds a threshold value or an optimization space exists, generating an optimization route adjustment suggestion considering the safety margin and the navigation efficiency. According to the invention, an integrated deep learning model is adopted, so that the accuracy and reliability of ship trajectory prediction are remarkably improved; and risk assessment and intelligent route optimization can be actively carried out, and the navigation safety and the operation efficiency are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of ship navigation, and in particular to a method for intelligently predicting a ship's navigation trajectory. Background Art

[0002] Navigation safety and efficiency are core issues for the shipping industry. Traditional navigation relies primarily on pilot experience, radar, electronic nautical charts, and the Automatic Identification System (AIS). While AIS provides information on vessel movements, its predictive capabilities are limited and it is primarily used for situational awareness.

[0003] Existing route planning and collision avoidance systems, some of which rely on deterministic rules or relatively simple trajectory extrapolation models, are limited in their prediction accuracy and comprehensive risk assessment when dealing with complex and dynamic ocean environments and multi-vessel encounter scenarios. These systems may not fully integrate heterogeneous data such as historical ship behavior patterns, real-time environmental factors, and weather forecasts for in-depth analysis. Therefore, how to use advanced intelligent technologies to achieve more accurate ship trajectory predictions, comprehensive potential risk assessments, and intelligent route optimization recommendations to improve navigation safety margins and economic efficiency is a key issue and development direction that needs to be urgently addressed in the current field of maritime technology. Summary of the Invention

[0004] In order to solve the problem that the above-mentioned existing route planning and collision avoidance systems partially rely on deterministic rules or relatively simple trajectory extrapolation models, and have limitations in prediction accuracy and comprehensiveness of risk assessment when dealing with complex and dynamic marine environments and multi-ship encounter scenarios, the present application provides an intelligent prediction method for ship navigation trajectories.

[0005] This application provides a method for intelligently predicting a ship's navigation trajectory, comprising the following steps: Step S1, dynamically acquiring and integrating a variety of navigation-related data covering the ship's own real-time status, historical navigation patterns, and current and forecasted environmental factors; Step S2: constructing and updating in real time a dynamic feature vector capable of characterizing the spatiotemporal motion behavior and intention of the ship based on the acquired and fused multiple navigation-related data; Step S3: inputting the dynamic feature vector into a pre-trained integrated deep learning prediction model to generate a probabilistic prediction sequence of the ship's navigation trajectory; Step S4, based on the probabilistic prediction sequence, combined with electronic nautical chart data and traffic management rules, assessing the potential collision risk and deviation risk of the predicted trajectory, and outputting a comprehensive risk assessment index; Step S5, based on the comprehensive risk assessment index and the probabilistic prediction sequence, when it is judged that the comprehensive risk assessment index exceeds the preset safety threshold or the probability of any single risk exceeds its emergency threshold, or when it is judged that there is significant optimization space, trigger and generate an optimized route adjustment suggestion that takes into account both safety margin and navigation efficiency.

[0006] Furthermore, the step S1 includes: Step S101, collecting the ship's AIS data, GNSS position data, compass heading data, and odometer speed data in real time; Step S102, obtaining real-time ocean environment data of wind, waves, and currents as well as future short-term meteorological and hydrological forecast data; Step S103, retrieving stored historical navigation track data and corresponding environmental data of the ship or ships of the same type; Step S104 , pre-processing the collected and acquired source data; and dynamically fusing the pre-processed multiple navigation-related data using an adaptive weighted fusion algorithm.

[0007] Furthermore, the adaptive weighted fusion algorithm used in step S104 is calculated as follows: ,in, Represents the fused navigation related data; Indicates the Navigation-related data from various sources; Indicates the The dynamic weight of each data source is calculated as follows: , Indicates the Real-time quality assessment factors for each data source; Indicates the The correlation factor between the data source and the current prediction task; f() is a predefined weight calculation function; when calculating the dynamic weight When: If the real-time quality assessment factor If the correlation factor is lower than the preset first threshold, the dynamic weight is reset to zero or assigned a preset minimum value; If the value is lower than a preset second threshold, the value of the dynamic weight is reduced accordingly.

[0008] Furthermore, the step S2 includes: Step S201, extracting and combining real-time kinematic parameters from the plurality of navigation-related data, the real-time kinematic parameters including the current position, speed, heading, and heading change rate of the vessel; Step S202, calculating a trend characteristic reflecting possible recent movement changes of the ship based on the historical navigation data of the ship and the current and predicted environmental factors; Step S203: combining the preset navigation plan of the ship or the conventional navigation pattern inferred by analyzing the historical navigation trajectory, quantifying the intention-indicative features that characterize the navigation target or trend of the ship; In step S204 , the instantaneous kinematic parameters, the trend characteristics, and the intention-indicative characteristics are combined into a feature vector as the dynamic feature vector, and the dynamic feature vector is reconstructed in real time according to the updates of the multiple navigation-related data.

[0009] Furthermore, the step S3 includes: Step S301, the integrated deep learning prediction model at least includes a temporal dependency capture module based on a long short-term memory network; and a feature enhancement module based on an attention mechanism; Step S302: The probabilistic prediction sequence includes the predicted position distribution within T time steps in the future. The predicted location Obey two-dimensional probability distribution ; Step S303: The model predicts the parameters of the two-dimensional probability distribution by calculation. and .

[0010] Furthermore, the two-dimensional probability distribution parameters are calculated in step S303 as follows: ,in represents the dynamic feature vector at the current time t constructed in step S2; The encoder network of the representation model; The decoder network of the representation model; represents the hidden state of the decoder network at the previous prediction time step t+k−1; represents the mean vector of a two-dimensional probability distribution; Represents a two-dimensional probability distribution The covariance matrix of Then make a judgment, if the covariance matrix If the determinant value of exceeds the preset uncertainty threshold, the time step is marked The prediction result is low confidence.

[0011] Furthermore, in step S301, the integrated deep learning prediction model includes a timing dependency capture module, which adopts a long short-term memory network structure to learn the time dependency pattern from the input dynamic feature vector sequence and generate a code with time context information; a feature enhancement module, which adopts an attention mechanism to assign importance weights to different parts of the code or the input dynamic feature vector sequence according to current prediction requirements, and aggregate the weighted parts to generate an enhanced feature representation, which is used for subsequent ship navigation trajectory prediction.

[0012] Furthermore, the step S4 includes: Step S401, calculating the risk probability of running aground or entering a restricted area for the probability distribution at each time step t+k in the probabilistic prediction sequence; Step S402: assessing the collision risk probability in combination with the status of other ships; Step S403, calculating the yaw risk probability; comprehensively calculating the comprehensive risk assessment index by combining the risk probability of running aground or entering a restricted area, the collision risk probability and the yaw risk probability.

[0013] Furthermore, the step S5 includes: Step S501, constructing an optimization objective function, aiming to minimize the voyage time or fuel consumption while maximizing the safety margin; Step S502 , the optimization objective function takes into account the predicted energy consumption cost, time cost, and safety-related penalty items; Step S503: using a heuristic search algorithm or a model-based optimization control method to solve a heading and speed adjustment sequence that optimizes the optimization objective function as the optimization route adjustment suggestion.

[0014] Furthermore, after generating the optimized route adjustment suggestion, it also includes: judging whether the increase in energy consumption or the predicted increase in time predicted according to the optimized route adjustment suggestion exceeds the preset economic constraint; if so, prompting the user or adjusting the strategy; and judging whether the safety penalty item after the predicted path optimization corresponding to the optimized route adjustment suggestion is still higher than the preset acceptable safety threshold; if so, generating a high-level alarm or recommending manual intervention.

[0015] The beneficial effects of the present invention compared with the prior art are: 1. This method integrates a vessel's real-time status, historical navigation patterns, and multi-dimensional environmental factors. It dynamically integrates multi-source navigation data through an adaptive weighted fusion algorithm to construct a dynamic feature vector reflecting the vessel's spatiotemporal motion and intentions. This method, combined with an integrated deep learning prediction model, effectively improves the accuracy and robustness of trajectory prediction. This method supports high-confidence probabilistic prediction outputs, accurately identifying potential yaw and collision risks in the future. This enhances the system's perception and response to complex sea conditions and changing navigation scenarios, significantly outperforming traditional static prediction methods based on rules or simple models.

[0016] 2. This invention incorporates risk quantification and multi-objective optimization mechanisms, combining electronic nautical charts with traffic regulations to assess the navigation safety of predicted routes. When risk exceeds a threshold or optimization potential exists, it automatically generates adjustment recommendations that balance safety and efficiency. This highly intelligent and practical approach surpasses existing technologies in predictive accuracy and risk identification. It also achieves the coordinated processing of automatic route optimization and energy consumption constraint control, providing a systematic solution for autonomous ship decision-making and intelligent navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flowchart of an exemplary step of the prediction method of the present invention.

[0018] Figure 2 The present invention is a flowchart of an exemplary step of obtaining navigation-related data.

[0019] Figure 3 A flowchart of an exemplary step of obtaining a dynamic vector according to the present invention.

[0020] Figure 4 A flow chart of exemplary steps for calculating a probabilistic prediction sequence according to the present invention.

[0021] Figure 5 A flowchart of an exemplary step for outputting a comprehensive risk assessment indicator according to the present invention.

[0022] Figure 6 A flow chart of exemplary steps for optimizing route adjustment recommendations according to the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the ship navigation trajectory intelligent prediction method provided by the embodiment includes the following steps: Step S1, dynamically acquiring and fusing various navigation related data covering the real-time state of the ship itself, historical navigation mode, and current and predicted environmental factors. The core is to build a comprehensive, multi-dimensional and dynamically updated data base, which is the premise of realizing accurate trajectory prediction and intelligent decision-making. The purpose is to overcome the possible information one-sidedness, inaccuracy or time lag of a single data source, and to provide as complete and reliable input as possible for subsequent feature extraction, model training and risk assessment by integrating heterogeneous data from different sensors, historical records and external prediction systems.

[0025] Figure 2 For an exemplary step flowchart of acquiring navigation related data in step S1 of the embodiment, it includes, Step S101, real-time collection of AIS data, GNSS position data, compass heading data, and log speed data of the ship. In the embodiment, the following data is acquired in real time by shipboard sensors and devices: AIS data, including MMSI, call sign, ship name, ship type, length, width, draft, current position, ground speed, ground heading, heading, turning rate, navigation state, destination, estimated time of arrival, etc. of the ship. AIS data is the core information of the ship's identity and dynamics. GNSS position data, such as high-precision ship latitude and longitude coordinates, altitude, and time stamp provided by GPS, GLONASS, Beidou, Galileo, etc. GNSS data usually has higher update frequency and higher accuracy than position data in AIS. Compass heading data: true heading or compass heading provided by the ship's compass. Log speed data: ship speed against water provided by the log.

[0026] Step S102, acquiring real-time ocean environmental data of wind, wave, and current, as well as future short-term meteorological and hydrological forecast data. Real-time wind speed, wind direction, significant wave height, wave period, wave direction, current speed, and current direction data at the current position of the ship are acquired through shipboard meteorological and hydrological sensors. At the same time, through the data interface with the shore-based meteorological and hydrological service agency, the meteorological and hydrological forecast data of the navigation area within a certain period of time are acquired, including the forecasted wind field, wave field, and current field information.

[0027] Step S103, retrieving stored historical navigation trajectory data and corresponding environmental data of the ship or similar type ships. From the shipboard or shore-based navigation data recording system, the trajectory data sequence of the ship's past navigation and the corresponding environmental data at that time are retrieved. If the historical data of the ship is insufficient, the historical data of similar type ships with similar parameters such as ship type, tonnage, and main engine power can be retrieved as reference. These data are helpful to learn the maneuvering characteristics and habitual route of the ship under different environments.

[0028] Step S104 , pre-processing the collected and acquired source data; and dynamically fusing the pre-processed multiple navigation-related data using an adaptive weighted fusion algorithm.

[0029] Preprocessing includes data cleaning to remove outliers and noise. Data interpolation and synchronization: Different sensors may collect data at different frequencies, requiring time alignment and necessary data interpolation to create a unified time series. Coordinate conversion and unit unification: Coordinate data from different sources is unified into a standard geographic coordinate system, with standardized units for speed, distance, and other data. Data validation and consistency checks, for example, compare AIS positions with GNSS positions, and SOG and speed log speeds with current information.

[0030] The adaptive weighted fusion algorithm used in step S104 is calculated as follows: ,in, Represents the fused navigation related data; Indicates the Navigation-related data from various sources; Indicates the The dynamic weight of a data source, which directly determines the The proportion of the data source in the final fusion result; the greater the weight, the greater the impact of its information on the fusion result. The calculation method is: , Indicates the Real-time quality assessment factors for each data source; Indicates the The correlation factor between the data source and the current prediction task; f() is a predefined weight calculation function; it indicates that the weight is and Jointly determined, when calculating dynamic weights When: If the real-time quality assessment factor If the correlation factor is lower than the preset first threshold, the dynamic weight is reset to zero or assigned a preset minimum value; If the value is lower than the preset second threshold, the value of the dynamic weight is reduced accordingly.

[0031] In step S2, based on the acquired and fused navigation data, a dynamic feature vector is constructed and updated in real time to characterize the vessel's spatiotemporal motion and intentions. The key point is that while the quality of the original fused data is improved, its form and dimensionality may not be optimal. Feature engineering is required to extract deeper information that is more discriminative and representative for the prediction model.

[0032] Figure 3This is a flowchart of an exemplary step of obtaining a dynamic vector in step S2 of this embodiment, including: Step S201, extracting and combining real-time kinematic parameters from a variety of navigation-related data, the real-time kinematic parameters including the current position, speed, heading, and heading change rate of the vessel; Step S202, calculating trend characteristics reflecting possible recent movement changes of the ship based on the ship's historical navigation data and current and predicted environmental factors; Step S203: combining the preset navigation plan of the ship or the conventional navigation pattern inferred by analyzing the historical navigation trajectory, quantifying the intention-indicative features that characterize the navigation target or trend of the ship; In step S204 , the real-time kinematic parameters, trend features, and intention-indicative features are combined into a feature vector as a dynamic feature vector, and the dynamic feature vector is reconstructed in real time according to updates of various navigation-related data.

[0033] The dynamic feature vector is constructed at time t and may include the following categories of features: ship status features, fused longitude, latitude, SOG, COG, HDG, ROT, draft, load percentage, main engine speed / power percentage, rudder angle, etc. Historical motion pattern features, trajectory point sequence over the past period, average SOG / COG, turning frequency, trajectory curvature, etc. Environmental impact features, current and predicted relative wind speed / direction, relative wave height / wave direction, relative current speed / direction, visibility, water depth, etc. Navigation intention and constraint features, information on the next waypoint of the planned route, relative relationship with the VTS reporting line or traffic separation system boundary, information on nearby obstructions, etc. The construction of the feature vector may involve normalization / standardization, scaling features of different dimensions to similar numerical ranges.

[0034] In step S3, the dynamic feature vector is input into a pre-trained integrated deep learning prediction model to generate a probabilistic prediction sequence of the ship's navigation trajectory.

[0035] Figure 4 This is an exemplary flowchart of the steps for calculating the probabilistic prediction sequence in step S3 of this embodiment, including: Step S301: The integrated deep learning prediction model includes at least a temporal dependency capture module based on a long short-term memory network and a feature enhancement module based on an attention mechanism. Step S302: The probabilistic prediction sequence includes the predicted position distribution within T time steps in the future. The predicted location Obeying a two-dimensional probability distribution ; Step S303: The model calculates and predicts the parameters of the above two-dimensional probability distribution and .

[0036] In step S303, the two-dimensional probability distribution parameters are calculated as follows: ,in represents the dynamic feature vector at the current time t constructed in step S2; The encoder network of the representation model; The decoder network of the representation model; represents the hidden state of the decoder network at the previous prediction time step t+k−1; represents the mean vector of a two-dimensional probability distribution; Represents a two-dimensional probability distribution The covariance matrix of Then make a judgment, if the covariance matrix If the determinant value of exceeds the preset uncertainty threshold, the time step is marked The prediction result is low confidence.

[0037] The principle of the Encoder-Decoder architecture is to map the input sequence to the output sequence. The Encoder is responsible for compressing the input information into an information-rich context representation, and the Decoder gradually generates a prediction sequence based on this context representation and its own previous state. Its purpose is to achieve the conversion from the current state to the future multi-step probabilistic prediction, where the mean vector represents the most likely location, and the covariance matrix The uncertainty range and correlation around the predicted position are quantified. Then make a judgment, if the covariance matrix If the determinant value of exceeds the preset uncertainty threshold, the prediction result of time step t+k is marked as low confidence.

[0038] In step S4, based on the probabilistic prediction sequence, combined with electronic nautical chart data and traffic management regulations, the potential collision and deviation risks of the predicted trajectory are assessed, and a comprehensive risk assessment index is output. The key is to interpret the abstract probabilistic trajectory prediction results generated by the model within the constraints of the real geographical environment and traffic regulations, thereby quantifying the specific navigation risks they may cause.

[0039] In step S301, the integrated deep learning prediction model includes a temporal dependency capture module, which adopts a long short-term memory network structure to learn the temporal dependency pattern from the input dynamic feature vector sequence and generate a code with temporal context information; a feature enhancement module, which adopts an attention mechanism to assign importance weights to different parts of the code or input dynamic feature vector sequence according to the current prediction requirements, and aggregate the weighted parts to generate an enhanced feature representation, which is used for subsequent ship navigation trajectory prediction.

[0040] Figure 5 An exemplary flowchart of the steps for outputting the comprehensive risk assessment index in step S4 of this embodiment includes: Step S401 calculates the probability of a grounding or restricted zone risk based on the probability distribution at each time step t+k in the probabilistic prediction sequence. The grounding risk calculates the probability of the vessel entering an area with a depth less than the safe draft at time t+k. This can be calculated by integrating the predicted position distribution with the depth contour. The restricted zone risk calculates the probability of the vessel entering the restricted zone at time t+k.

[0041] Step S402: Evaluate the collision risk probability in combination with the status of other ships.

[0042] Obtain real-time AIS data from other nearby ships and perform trajectory predictions for them. For each predicted position distribution of our ship and the predicted position distribution of the other ship, calculate the probability of a collision between our ship and the other ship j at time t+k. This typically involves calculating the probability distributions of the closest approach distance and closest approach time of the two ships at time t+k, or directly calculating the probability that the predicted position distributions of the two ships overlap and are less than the safe distance. For example, consider the hull models of the two ships and calculate the probability that these two geometries overlap at time t+k.

[0043] Step S403, calculating the risk probability of deviation; comprehensively calculating the risk probability of running aground or entering a restricted area, the risk probability of collision and the risk probability of deviation, and calculating a comprehensive risk assessment index.

[0044] Deviation risk: If there is a pre-defined planned route or channel centerline, calculate the probability that the predicted trajectory will significantly deviate from the planned route or channel boundary. The comprehensive risk assessment indicator can be the maximum value of each risk probability, a weighted sum, or a more complex assessment based on a risk matrix.

[0045] Step S5, based on the comprehensive risk assessment indicators and probabilistic prediction sequence, triggers and generates optimized route adjustment recommendations that balance safety margins and navigation efficiency when the comprehensive risk assessment indicators exceed the preset safety threshold, the probability of any individual risk exceeds its emergency threshold, or when significant room for optimization is determined. The core principle is that when the system predicts unacceptable future risks or potential for route efficiency improvements, it uses optimization algorithms to proactively search for and recommend adjustment strategies that effectively mitigate risks or improve navigation economy.

[0046] Figure 6 This is an exemplary flowchart of the steps for optimizing route adjustment suggestions in step S5 of this embodiment, including: Step S501: construct an optimization objective function, aiming to minimize the voyage time or fuel consumption while maximizing the safety margin. It can be expressed as: ,in It can be taken as the total sailing time of the predicted route, or the total fuel consumption can be estimated based on the ship's speed-fuel consumption curve and the predicted speed series. Can be defined as a comprehensive risk assessment indicator or penalty items related to the probability of each risk, such as ,in It is a monotonically increasing function. The higher the risk, the greater the penalty. It can also be defined as the degree of deviation from the minimum safe distance.

[0047] Step S502 : Optimizing the objective function by taking into account the predicted energy consumption cost, time cost, and safety-related penalty items.

[0048] In step S503, a heuristic search algorithm or a model-based optimization control method is used to determine the heading and speed adjustment sequence that optimizes the optimization objective function, as a recommendation for optimizing the route adjustment. Heuristic search discretizes the possible heading and speed adjustments and searches for an adjustment sequence within a future time period that minimizes the objective function J under the predicted trajectory. Model predictive control, within each control cycle, determines the optimal control sequence within a finite future time domain based on the current state and the prediction model, executes the first control action of the sequence, and repeats this process in the next cycle. The optimal control sequence can handle constraints and has feedforward capabilities.

[0049] After generating the optimized route adjustment suggestion, determine whether the predicted increase in energy consumption or predicted increase in time based on the optimized route adjustment suggestion exceeds the preset economic constraint. If so, prompt the user or adjust the strategy; and determine whether the safety penalty item after optimization of the predicted path corresponding to the optimized route adjustment suggestion is still higher than the preset acceptable safety threshold. If so, generate a high-level alarm or recommend manual intervention.

[0050] The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in similar ways. As long as they do not deviate from the scope defined by the invention, they should all fall within the scope of protection of the present invention.

Claims

1. A method for intelligently predicting a ship's navigation trajectory, characterized by: The following steps are included: Step S1, dynamically acquiring and integrating a variety of navigation-related data covering the ship's own real-time status, historical navigation patterns, and current and forecasted environmental factors; Step S2: constructing and updating in real time a dynamic feature vector capable of characterizing the spatiotemporal motion behavior and intention of the ship based on the acquired and fused multiple navigation-related data; Step S3: inputting the dynamic feature vector into a pre-trained integrated deep learning prediction model to generate a probabilistic prediction sequence of the ship's navigation trajectory; Step S4, based on the probabilistic prediction sequence, combined with electronic nautical chart data and traffic management rules, assessing the potential collision risk and deviation risk of the predicted trajectory, and outputting a comprehensive risk assessment index; Step S5, based on the comprehensive risk assessment index and the probabilistic prediction sequence, when it is judged that the comprehensive risk assessment index exceeds the preset safety threshold or the probability of any single risk exceeds its emergency threshold, or when it is judged that there is significant optimization space, trigger and generate an optimized route adjustment suggestion that takes into account both safety margin and navigation efficiency.

2. The method for intelligently predicting a ship's navigation trajectory according to claim 1, characterized in that: The step S1 includes: Step S101, collecting the ship's AIS data, GNSS position data, compass heading data, and odometer speed data in real time; Step S102, obtaining real-time ocean environment data of wind, waves, and currents as well as future short-term meteorological and hydrological forecast data; Step S103, retrieving stored historical navigation track data and corresponding environmental data of the ship or ships of the same type; Step S104 , pre-processing the collected and acquired source data; and dynamically fusing the pre-processed multiple navigation-related data using an adaptive weighted fusion algorithm.

3. The method for intelligently predicting a ship's navigation trajectory according to claim 2, characterized in that: The adaptive weighted fusion algorithm used in step S104 is calculated as follows: ,in, Represents the fused navigation related data; Indicates the Navigation-related data from various sources; Indicates the The dynamic weight of each data source is calculated as follows: , Indicates the Real-time quality assessment factors for each data source; Indicates the The correlation factor between the data source and the current prediction task; f() is a predefined weight calculation function; when calculating the dynamic weight When: If the real-time quality assessment factor If the correlation factor is lower than the preset first threshold, the dynamic weight is reset to zero or assigned a preset minimum value; If the value is lower than a preset second threshold, the value of the dynamic weight is reduced accordingly.

4. The method for intelligently predicting a ship's navigation trajectory according to claim 1, characterized in that: The step S2 includes: Step S201, extracting and combining real-time kinematic parameters from the plurality of navigation-related data, the real-time kinematic parameters including the current position, speed, heading, and heading change rate of the vessel; Step S202, calculating a trend characteristic reflecting possible recent movement changes of the ship based on the historical navigation data of the ship and the current and predicted environmental factors; Step S203: combining the preset navigation plan of the ship or the conventional navigation pattern inferred by analyzing the historical navigation trajectory, quantifying the intention-indicative features that characterize the navigation target or trend of the ship; In step S204 , the instantaneous kinematic parameters, the trend characteristics, and the intention-indicative characteristics are combined into a feature vector as the dynamic feature vector, and the dynamic feature vector is reconstructed in real time according to the updates of the multiple navigation-related data.

5. The method for intelligently predicting a ship's navigation trajectory according to claim 1, characterized in that: The step S3 includes: Step S301, the integrated deep learning prediction model at least includes a temporal dependency capture module based on a long short-term memory network; And a feature enhancement module based on the attention mechanism; Step S302: The probabilistic prediction sequence includes the predicted position distribution within T time steps in the future. The predicted location Obeying a two-dimensional probability distribution ; Step S303: The model predicts the parameters of the two-dimensional probability distribution by calculation. and .

6. The method for intelligently predicting a ship's navigation trajectory according to claim 5, characterized in that: In step S301, the integrated deep learning prediction model includes a temporal dependency capture module, which adopts a long short-term memory network structure to learn a temporal dependency pattern from the input dynamic feature vector sequence and generate a code with temporal context information; A feature enhancement module employs an attention mechanism to assign importance weights to different parts of the encoding or the input dynamic feature vector sequence according to current prediction requirements, and aggregates the weighted parts to generate an enhanced feature representation, which is used for subsequent ship navigation trajectory prediction.

7. The method for intelligently predicting a ship's navigation trajectory according to claim 5, characterized in that: The two-dimensional probability distribution parameters are calculated in step S303 as follows: ,in represents the dynamic feature vector at the current time t constructed in step S2; The encoder network of the representation model; The decoder network of the representation model; represents the hidden state of the decoder network at the previous prediction time step t+k−1; represents the mean vector of a two-dimensional probability distribution; Represents a two-dimensional probability distribution The covariance matrix of Then make a judgment, if the covariance matrix If the determinant value of exceeds the preset uncertainty threshold, the time step is marked The prediction result is low confidence.

8. The method for intelligently predicting a ship's navigation trajectory according to claim 1, characterized in that: The step S4 comprises: Step S401, calculating the risk probability of running aground or entering a restricted area for the probability distribution at each time step t+k in the probabilistic prediction sequence; Step S402: assessing the collision risk probability in combination with the status of other ships; Step S403, calculating the yaw risk probability; comprehensively calculating the comprehensive risk assessment index by combining the risk probability of running aground or entering a restricted area, the collision risk probability and the yaw risk probability.

9. The method for intelligently predicting a ship's navigation trajectory according to claim 1, characterized in that: The step S5 includes: Step S501, constructing an optimization objective function, aiming to minimize the voyage time or fuel consumption while maximizing the safety margin; Step S502 , the optimization objective function takes into account the predicted energy consumption cost, time cost, and safety-related penalty items; Step S503: using a heuristic search algorithm or a model-based optimization control method to solve a heading and speed adjustment sequence that optimizes the optimization objective function as the optimization route adjustment suggestion.

10. The method for intelligently predicting a ship's navigation trajectory according to claim 9, characterized in that: After generating the optimized route adjustment suggestion, it also includes: judging whether the increase in energy consumption or the predicted increase in time predicted according to the optimized route adjustment suggestion exceeds the preset economic constraint; if so, prompting the user or adjusting the strategy; and judging whether the safety penalty item after the predicted path optimization corresponding to the optimized route adjustment suggestion is still higher than the preset acceptable safety threshold; if so, generating a high-level alarm or recommending manual intervention.

Images