Ship traffic intelligent prediction and cooperative guidance method and system based on NAVDAT communication

Abstract

Disclosed is a ship traffic intelligent prediction and cooperative guidance method and system based on NAVDAT communication: the method comprises the following steps: 1) collecting NAVDAT broadcast data, and processing the collected data to obtain a unified format tensor; 2) traffic state modeling and prediction, including dynamic prediction of traffic density, berthing waiting time and potential congestion points in a specific water area based on a graph neural network and a sliding time window mechanism, and fusion of ship behavior, task attribute and environmental factors; 3) according to the credibility level of the prediction result, constructing a differentiated information broadcast content package, and broadcasting and pushing the recommended strategy to multiple ships through the NAVDAT system to realize consistent guidance; 4) optimizing ship response feedback, obtaining the response behavior of the ship to the suggestion through ship-shore communication or indirect data collection, using the execution deviation data to optimize the prediction update and pushing mechanism, and constructing a prediction-guidance-feedback-learning closed-loop optimization path.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a method and system for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication. Background Technology

[0002] Maritime and port vessel traffic management relies on wireless communication management systems, but current maritime traffic management systems still have some shortcomings. For example, most current maritime traffic management systems collect vessel information based on AIS (Automatic Identification System), including static or dynamic attributes such as vessel position, speed, and heading, and use time series prediction models (such as LSTM) to make short-term predictions of vessel paths. This approach can support functions such as vessel path tracking and berthing warnings, but its prediction accuracy and response speed are limited due to the low data transmission rate of AIS, its susceptibility to channel congestion, and its lack of support for complex structured information transmission. Some ports have introduced static rule models, which prioritize vessels based on predetermined berth plans and ETA (Estimated Time of Arrival), such as first-come, first-served or priority berthing. These systems lack real-time environmental perception and prediction capabilities, and lack dynamic adjustment mechanisms in the face of emergencies (such as weather changes or traffic anomalies), easily leading to resource waste and increased congestion. VDES (VHF Data Exchange System), as an extension of AIS, supports two-way data communication and can realize information exchange between ships and shore. Some existing studies have attempted to push scheduling information and waterway status to ship terminals through VDES. However, since VDES is still in the early stages of deployment, the terminal penetration rate is not high, the system architecture is complex, and it is still mainly based on point-to-point communication, it is difficult to achieve broadcast-style consistent situational awareness.

[0003] Therefore, new methods and technologies are needed to at least partially overcome the problems existing in the prior art. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method and system for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication. Utilizing the structured data format, highly reliable redundant transmission mechanism, and regional synchronous broadcasting characteristics of NAVDAT broadcasting, it achieves accurate prediction, hierarchical guidance, and a closed loop of behavioral feedback for maritime ship traffic conditions. The system comprises a structured information parsing module, a traffic condition modeling and prediction module, a confidence-based hierarchical guidance module, and a ship response feedback optimization module. These modules work closely together in data input, model inference, information dissemination, and behavioral feedback.

[0005] According to one aspect of the present invention, a method for intelligent prediction and cooperative guidance of ship traffic based on NAVDAT communication is provided, comprising:

[0006] 1) Collect NAVDAT broadcast data and process the collected data to obtain a uniformly formatted tensor. Where r is the length of the time window, and d * This represents the number of effective feature dimensions after processing.

[0007] 2) Based on the processed data from step 1), perform traffic state modeling and prediction, including:

[0008] 2.1) Construct a heterogeneous graph structure based on all target ships within the current time window. The node set V contains all individual ships that appear within the sliding window [t-r+1,t]. Each node... Each node corresponds to a single ship; the edge set ε represents the spatial proximity, similar task relationships, or port scheduling correlations between ships; the attribute vector of each node is defined as the feature of its structured state at time t: Where, d t (i) Let be the standardized structured vector of the i-th ship at time t;

[0009] To quantify the boundary rights relationships between ships, two nodes v are defined. i With v j The edge weight is e ij Its calculation is based on the spatial distance between the two ships at time t, mission consistency, and environmental similarity, and is defined as:

[0010]

[0011] Where p i =(x i y i ), p j =(x j y j ) are the position coordinates of the i-th and j-th ships, respectively, and σ is the distance scale adjustment factor. This is a task type consistency indicator function, taking the value 1 or 0. env (i, j) represents the cosine similarity based on the meteorological tide field, and α, β, γ are the weighting factors of the three types of relationships;

[0012] 2.2) A graph neural network based on an attention mechanism is used for node state updates and feature fusion. Information aggregation is completed with the help of a graph attention network (GAT) to obtain the global embedding representation h of each ship. i (L) ,

[0013] 2.3) A multi-task prediction head is used to jointly predict the future state of each ship, and the output vector includes the position offset Δp. i Velocity change Δv iberth occupancy probability q i berth and the task transition probability distribution q i task The prediction formula is as follows:

[0014]

[0015] in W p With b p These are the prediction layer weights and biases, respectively.

[0016] 3) Transform the future situation information obtained in step 2) into behavioral guidance instructions for multiple targets and scenarios, including:

[0017] 3.1) Evaluate the confidence level of the prediction output in step 2) to obtain the overall confidence level S of the current prediction for the i-th ship. i :

[0018] 3.2) Based on the confidence index S mentioned above i Construct a hierarchical structure for the bootstrapping strategy: when s i ≥τ H When τ, the system generates direct control boot instructions; when τ L ≤s i <τ H When s i <τ L At that time, the system only generates passive broadcast alerts, where τ H τ L These are the high and low thresholds, respectively.

[0019] 4) Optimize ship response feedback, including simultaneously broadcasting a unified feedback format and return window to all target ships by embedding feedback response definition fields while issuing prediction and guidance information, thereby achieving standardized collection of multi-dimensional behavioral feedback and providing basic data support for feedback-driven model optimization and guidance scheduling.

[0020] According to an embodiment of the present invention, in step 1), the received NAVDAT broadcast data includes the following structured fields: ship dynamic information, environmental status information, and mission and scheduling information. The ship dynamic information includes geographic coordinates, speed, and heading, and the environmental status information includes weather codes. t and tidal level c t Task and scheduling information includes estimated berthing time T t ETA Berth number b t Task type m t .

[0021] According to an embodiment of the present invention, in step 1), data processing includes:

[0022] 1.1) Encode the field at time t into a unified structured feature vector:

[0023]

[0024] Where, p t =(x t y t ), indicating the ship's position; v t θt is the speed; θt is the heading angle; d is the total feature dimension.

[0025] 1.2) Construct the input sequence based on the sliding time window mechanism to form a time-consistent dataset. Let the time window length be r, then the input window at time t is:

[0026]

[0027] 1.3) Perform missing data completion, including:

[0028] Suppose that the nth field within the sliding window has a missing value at time point k. Its complete estimate is defined as:

[0029]

[0030] Where, μ n σ n Let be the mean and standard deviation of field n, respectively; λ∈[0,1] be the offset adjustment factor; and ε be the smoothing term. The original value of field n at time points other than k;

[0031] 1.4) After completion, perform standardization processing on all numeric fields, including:

[0032] Let μ be the mean of the g-th field in the completed structured feature vector within the window. g The standard deviation is σ g The standardized value is:

[0033] in Let g be the original value of the g-th field at time t.

[0034] After standardization, a data matrix with zero mean and unit variance is obtained:

[0035] The completed and standardized sliding window sequence is denoted as W. t * ,Right now:

[0036] Then let d * ≤d, we obtain a unified tensor:

[0037] According to an embodiment of the present invention, step 2.2) includes:

[0038] Update node v i The representation at layer l is:

[0039]

[0040] in W represents the set of neighbors connected to node i. (l) Let be the linear transformation matrix of the l-th layer, and σ(·) represent the activation function; Attention coefficient;

[0041] The attention coefficient reflects the degree of information contribution of node j to node i, and is calculated as follows:

[0042]

[0043] Where α is the learnable attention vector, and || denotes the vector concatenation operation;

[0044] After L-layer aggregation, a global embedding representation of each ship is obtained. This statement incorporates its timing status, the influence of neighboring vessels, and the scheduling environment.

[0045] According to an embodiment of the present invention, step 3.1) includes:

[0046] Let the prediction result for the i-th ship at time t be:

[0047]

[0048] in, Let Δv be the future position offset vector. i For velocity change scalar, This represents the probability of berth occupancy. The predicted distribution for task categories;

[0049] An anomaly measure based on Mahalanobis distance quantifies the confidence level of each predicted value. Let the mean vector of the output be μ. y The covariance matrix is ​​Σ y Then the overall confidence level of the current prediction for the i-th ship is:

[0050]

[0051] Confidence score s i ∈(0,1], the closer the value is to 1, the more reliable the prediction result is.

[0052] According to an embodiment of the present invention, step 3.2) includes combining decision weights R. i Provide guidance:

[0053] Where ω1, ω2, ω3 are normalized weighting coefficients. For the probability of future berth occupancy conflicts, For the target path density, The urgency level of the current mission; for high-confidence vessels (i.e., s) i ≥τ H And R i For vessels exceeding a given threshold, the system will issue detailed guidance instructions via NAVDAT structured communication broadcast frames, including recommended speed, berth sequence adjustments, and port entry waiting strategies.

[0054] The probability of berth conflict is calculated as follows:

[0055] sim represents the predicted berthing time for the i-th ship. T (·) represents the time period overlap function;

[0056] The calculation of historical track clustering within a region is defined as follows:

[0057] in, For p i Let r be the set of track points in the spatial neighborhood of the center radius r, and δ be the length of the prediction time window;

[0058] Extract directly based on task type weight or scheduling level field.

[0059] According to an embodiment of the present invention, step 3) further includes introducing a prediction execution feedback comparison mechanism to recalculate the confidence level s. i upd ,

[0060] Let the actual behavior of the ship after execution be... The original predicted location was The error rate is then calculated as follows:

[0061] Where ε is a very small positive number to prevent division by zero;

[0062] When a ship is in K consecutive predictions ∈ i >∈max If the behavior deviates from the current distribution model, the confidence level will be automatically recalculated, and the ship's most recent M prediction results will be retrieved. Calculate the local mean and covariance:

[0063]

[0064] Then, using the fusion factor λ and weighting, we obtain:

[0065]

[0066] Finally, the confidence level s is recalculated. i upd :

[0067]

[0068] According to an embodiment of the present invention, step 4) includes:

[0069] 4.1) The feedback content from the ship within the specified response window [t, t+δ] uniformly constitutes the feedback behavior vector. This vector will be compared with the prediction guidance vector originally released by the system. Perform a structured comparison to generate a response bias vector: And further calculate the standardized deviation rate index:

[0070]

[0071] Where ||·||2 represents the Euclidean norm, ε>0, and is the smoothing regularization factor;

[0072] 4.2) Construct a structured response sample dataset based on step 4.1): Where c i (t) This is the feature vector of the ship's environmental context at that time;

[0073] 4.3) Based on this dataset, train the feedback bias scoring function f. score Used to evaluate the expected response error under specific forecast and environment:

[0074]

[0075] 4.4) For any predicted guiding vector Calculate its risk of deviation using a scoring function:

[0076] Based on the set feedback robustness expectation, the release command is modulated as follows:

[0077] Where α∈[0,1] is the modulation coefficient.

[0078] According to an embodiment of the present invention, step 4) further includes 4.5), constructing a response deviation space based on structured feedback. Density clustering is used to identify behavioral response clusters with common deviation characteristics, and these clusters are assigned behavioral labels φ. k This forms a set of response exception labels A.

[0079] According to another aspect of the present invention, a ship traffic intelligent prediction and cooperative guidance system based on NAVDAT communication is provided, comprising:

[0080] The structured information parsing module is used to collect NAVDAT broadcast data and process the collected data to obtain a unified format tensor. Where r is the length of the time window, and d * This represents the number of effective feature dimensions after processing.

[0081] The traffic condition modeling and prediction module is used to perform traffic condition modeling and prediction based on the data processed by the structured information parsing module.

[0082] The confidence-based hierarchical guidance module is used to transform the obtained future situation information into behavioral guidance instructions for multiple targets and scenarios;

[0083] The ship response feedback optimization module is used to optimize ship response feedback. This includes simultaneously broadcasting a unified feedback format and return window to all target ships by embedding feedback response definition fields while issuing prediction and guidance information. This enables standardized collection of multi-dimensional behavioral feedback and provides basic data support for feedback-driven model optimization and guidance scheduling.

[0084] The intelligent prediction and collaborative guidance method and system for ship traffic based on NAVDAT communication according to the embodiments of the present invention can achieve beneficial technical effects:

[0085] 1. Improved communication capacity and broadcast consistency, enabling high-density information synchronization.

[0086] Existing technologies commonly use AIS systems for ship-to-shore data transmission. However, the AIS communication channel rate is only 9.6 kbps, and the frequency band is congested, especially in high-density waters such as ports, which can easily lead to delays or loss of high-frequency information packets such as scheduling information, channel restrictions, and weather warnings. In contrast, the NAVDAT communication protocol upon which this invention is based uses a mid-band modulation method, with a downlink rate typically between 15 kbps and 25 kbps. It possesses stable medium-bandwidth transmission capabilities and supports structured data frame formats, enabling the unified packaging and broadcasting of various scheduling elements, prediction parameters, and feedback structures.

[0087] Through the structured information parsing module, the system can complete the synchronous distribution of high-density information to all ships in a short time, ensuring that multiple target ships in the port area have a unified perception of key navigation constraints and scheduling intentions. This solves the problem of "information asynchrony leading to behavioral conflicts" under existing technologies and provides a basic guarantee of information consistency for downstream predictive modeling and collaborative guidance.

[0088] 2. In terms of traffic condition modeling and prediction accuracy, the model's global adaptability has been improved.

[0089] Existing prediction systems mostly rely on the ship's own behavioral trajectory (such as speed, heading, and historical path) to construct time series models, neglecting the critical impact of environmental and scheduling factors on ship behavior. This leads to a sharp drop in prediction accuracy and a lack of robustness under special events (such as temporary scheduling changes or weather disturbances). This invention uses a structured information parsing module to transform factors such as tides, weather, and port scheduling carried in NAVDAT broadcasts into heterogeneous environmental inputs, and then introduces a three-dimensional dynamic graph representation of "ship-environment-task" into the traffic state modeling and prediction module.

[0090] By combining a graph neural network-based modeling strategy, the system can simultaneously capture the spatial dynamic relationships between ships and the external environment's dependent structure, effectively improving its global modeling capability for the overall traffic situation's evolution. Experiments demonstrate that under complex environmental disturbances, this system can significantly reduce the average prediction error and improve prediction stability in situations with multiple ships densely berthing, exhibiting a clear system-level accuracy advantage.

[0091] 3. Regarding the information push strategy, differentiated and tiered push notifications and misleading control have been implemented.

[0092] Existing systems typically distribute guidance suggestions via a uniform broadcast, failing to differentiate between high and low prediction confidence levels. This can easily lead to misleading results of "strong suggestions + low confidence" in areas of model uncertainty, especially in complex scenarios (such as collision avoidance and temporary task switching). This invention designs a guidance level classification mechanism based on response deviation rate in the confidence-level hierarchical guidance module. It utilizes historical execution deviations to construct a confidence function and performs graded evaluation of prediction results.

[0093] High-confidence predictions are broadcast directly as "strong recommendation" guidance commands via NAVDAT, while low-confidence results are only issued as advisory information. This controls the range of autonomous responses by the vessel, effectively avoiding misjudgments or operational conflicts caused by insufficient confidence. The module also features a dynamic confidence correction mechanism, ensuring the system continuously adjusts its push strategy based on feedback, improving the accuracy and reliability of information delivery.

[0094] 4. A closed-loop optimization mechanism was constructed to address system feedback capability and model evolution.

[0095] Traditional ship guidance systems are mostly one-time deployment structures, lacking mechanisms for collecting ship behavior data and model callbacks after responses. Even if prediction errors exist, they cannot be dynamically corrected. Furthermore, because communication mechanisms such as AIS cannot support multi-field structured feedback, feedback data acquisition is almost non-existent. This invention leverages the high-bandwidth structured communication capabilities of NAVDAT to embed feedback response definition segments into guidance information frames, achieving unified acquisition of multi-dimensional response behaviors (track deviation, berth success rate, task completion status, etc.). The feedback optimization module constructs a behavior deviation function based on structured feedback data and uses a scoring model f... score The system quantifies the expected deviations of prediction results under different scenarios and further constructs behavioral clustering labels to achieve adaptive evolution of the prediction model and strategy structure. This feedback-driven closed-loop structure of "prediction-guidance-feedback-optimization" is the first of its kind in the maritime communication scenario, solving the technical problems of "uncontrollable prediction failures and untrackable errors" in existing systems.

[0096] 5. Breakthroughs have been achieved in regional collaborative sensing and consistency of multi-ship scheduling.

[0097] Because AIS is a point-to-point communication structure, ships lack the ability to share overall traffic conditions and cannot perceive the dispatch responses of surrounding ships, which can easily lead to localized conflicts and waterway disputes. This invention, based on the wide-area broadcast characteristics of NAVDAT, constructs a unified predictive dispatch framework at the system layer. It synchronously distributes the prediction results, guidance strategies, and key behavioral markers of each ship to all target ships, enabling each ship to possess regional-level collaborative perception capabilities.

[0098] Based on this, vessels can dynamically adjust their behavior locally to avoid mutually exclusive paths, forming a collaborative avoidance mechanism based on shared predictions. This mechanism significantly improves the system's scheduling efficiency and safety margin in scenarios such as port congestion, anchorage queuing, and complex diversion, achieving a systemic leap from "local response" to "regional collaboration." Attached Figure Description

[0099] Figure 1 This is a flowchart illustrating the intelligent prediction and collaborative guidance method for ship traffic based on NAVDAT communication according to an embodiment of the present invention.

[0100] Figure 2 This is a system architecture diagram of intelligent prediction and cooperative guidance of ship traffic based on NAVDAT communication according to an embodiment of the present invention; and

[0101] Figure 3 This is a schematic diagram of the structure of a ship traffic intelligent prediction and collaborative guidance system based on NAVDAT communication according to an embodiment of the present invention. Detailed Implementation

[0102] The present invention can be better understood from the accompanying drawings and the following embodiments. However, those skilled in the art will readily understand that the descriptions in the embodiments are for illustrative purposes only and should not, and will not, limit the scope of the invention.

[0103] Figure 1 This is a flowchart illustrating the intelligent prediction and collaborative guidance method for ship traffic based on NAVDAT communication according to an embodiment of the present invention. Figure 2 The figure shows a system architecture diagram for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to an embodiment of the present invention. As shown in the figure, the method and system for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to the embodiment are as follows:

[0104] 1. Structured Information Parsing

[0105] In the proposed intelligent prediction and collaborative guidance system for ship traffic based on NAVDAT communication, the structured information parsing module serves as the system's data entry module, primarily responsible for receiving, parsing, normalizing, and standardizing NAVDAT broadcast frames. Because NAVDAT employs a medium-speed broadcast mechanism based on the MF / HF band, its data frames are characterized by clear structure, fixed fields, and rich content types. Therefore, this module can achieve processing results far exceeding those of traditional AIS systems in terms of semantic integrity and timeliness, providing high-quality time-series data input for subsequent graph neural network modeling.

[0106] The received NAVDAT broadcast frame may contain the following structured fields: ship dynamic information (such as geographic coordinates, speed, and heading), environmental status information (such as weather code Wt and tide class Ct), and mission and scheduling information (such as estimated berthing time T). t ETA Berth number b t Task type m t These fields are encoded into a unified structured feature vector at time t.

[0107]

[0108] Where, p t =(x t y t ) represents the ship's position, vt is the speed (knots), θt is the heading angle (degrees), and d is the total feature dimension.

[0109] Considering the fixed-period redundancy of NAVDAT broadcasts, a sliding time window mechanism can be used to construct the input sequence, forming a time-consistent, high-quality dataset. Let the time window length be r, then the input window at time t is:

[0110] This represents the sequence of historical states taken r steps forward from time t.

[0111] Considering that fields may be missing in actual broadcasts (such as T) t ETA b t To ensure the completeness of feature statistics, missing value completion can be performed before standardization. Suppose that a certain field in the nth column of the sliding window has missing values ​​at time point k. (that is, at time point k, p) t v t θ t w t c t , b t m t The completion estimate of the nth value (i.e., the nth field) in this sequence is defined as:

[0112]

[0113] Where, μ n σ n Here, λ represents the mean and standard deviation of field n, respectively; λ∈[0,1] is the offset adjustment factor; and ε is the smoothing term to prevent the denominator from being zero. This represents the original value of field n at time points other than k. This method combines the central tendency and dispersion of the field, making it suitable for robust completion in dynamically changing scenarios.

[0114] To avoid the impact of different units of measurement on model convergence, standardization can be performed on all numerical fields. Let the mean of the g-th field within the window be μ. g The standard deviation is σ g The standardized value is:

[0115]

[0116] in, The original value of the g-th field at time t, the standardized result. It satisfies the characteristics of zero mean and unit variance.

[0117] After standardizing all fields, a data matrix with zero mean and unit variance is obtained:

[0118]

[0119] The completed and standardized sliding window sequence is denoted as W. t * ,Right now:

[0120] Ultimately, let d *≤d represents the number of valid feature dimensions retained (which may have undergone missing feature removal, encoding expansion, etc.). The system organizes the standardized results into a uniform tensor:

[0121]

[0122] The tensor x t This is the final output of this module, which serves as the direct input for subsequent graph modeling and prediction modules. It possesses field consistency, time comparability, and structural completeness, providing solid data support for dynamic multi-ship collaborative modeling.

[0123] After normalization and completion, all structured fields will be merged into a uniform format tensor. Where d* represents the number of effective feature dimensions after processing. This tensor will serve as the input for subsequent modeling modules, offering key advantages such as temporal consistency, field completeness, and scale comparability.

[0124] Through the above mechanism, the structured information parsing module not only solves the problems of data sparsity and incomplete fields in the traditional AIS mode, but also combines the advantages of NAVDAT's broadcast structure to complete the task of constructing high-frequency, highly stable, and computable maritime multimodal time series data, laying a solid technical foundation for realizing global perception and predictive modeling.

[0125] 2. Traffic Condition Modeling and Prediction

[0126] After completing the structured information parsing module, the system has obtained a standardized multi-ship state data tensor organized in the form of a sliding time window. Where r represents the time window length, and d* represents the feature dimension after completion and standardization. To construct a ship traffic evolution model for dynamic environments, this invention designs a prediction method based on the fusion of graph structure modeling and attention mechanisms, fully exploring the spatiotemporal dependencies between ships, the correlation of port scheduling, and the impact of environmental state changes on traffic evolution. Finally, it outputs the predicted state vector of each ship at future moments, including its expected position, speed, berth occupancy status, and task switching probability, which is used by the collaborative guidance module to formulate an instruction strategy that optimizes overall regional efficiency.

[0127] First, a heterogeneous graph structure is constructed based on all target ships within the current time window. The node set v contains all individual ships that appear within the sliding window [t-r+1, t], and each node... Each node corresponds to a single ship; the edge set ε represents the spatial proximity, similar task relationships, or port scheduling correlations between ships. The attribute vector of each node is defined as the feature of its structured state at the most recent time t:

[0128] in Let be the standardized structured vector of the i-th ship at time t, initialized as the input node embedding of the graph neural network. To quantify the edge weight relationships between ships, let the edge weight between two nodes vi and vj be e. ij Its calculation is based on the spatial distance between the two ships at time t, mission consistency, and environmental similarity, and is defined as:

[0129]

[0130] Where p i =(x i y i ), p j =(x j y j ) are the position coordinates of the i-th and j-th ships, respectively. The task type consistency indicator function takes the value of 1 or 0, σ is the distance scale adjustment factor, and sim env (i, j) represents the cosine similarity based on the meteorological tide field, and α, β, γ are the weighting factors of the three types of relationships, which can be determined based on existing data or experience.

[0131] After establishing the graph structure, a graph neural network based on an attention mechanism is used for node state updates and feature fusion, and information aggregation is completed with the help of a graph attention network (GAT). The representation of node vi in ​​layer l is updated as follows:

[0132]

[0133] in W represents the set of neighbors connected to node i. (l) Let be the linear transformation matrix of the l-th layer, and σ(·) represent the activation function (e.g., leaky ReLU, ReLU). Attention coefficients. This reflects the degree of information contribution of node j to node i, and is calculated as follows:

[0134]

[0135] Where a is a learnable attention vector, and || denotes the vector concatenation operation.

[0136] After L-layer aggregation, a global embedding representation of each ship is obtained. This representation incorporates the vessel's temporal state, the influence of neighboring vessels, and the scheduling environment. Based on this, a multi-task prediction head is used to jointly predict the future state of each vessel, with the output vector including the position offset Δp. i Velocity change Δv i berth occupancy probability and task transfer probability distribution The prediction formula is as follows:

[0137]

[0138] in Wp and b p These are the prediction layer weights and biases, respectively.

[0139] This module achieves spatiotemporal modeling capabilities for complex port traffic situations by integrating standardized sliding window information generated by NAVDAT broadcasts with a multi-ship collaborative graph structure. The graph attention mechanism effectively uncovers heterogeneous connections between ships, enhancing the model's ability to perceive high-dimensional variables such as channel congestion and scheduling conflicts. This significantly outperforms traditional prediction methods based on single-ship time series, demonstrating stronger global coordination and dynamic adaptability, and providing accurate and reliable predictive input for the subsequent guidance instruction generation module.

[0140] 3. Confidence level stratification guidance

[0141] In the intelligent prediction and collaborative guidance system for ship traffic based on NAVDAT communication proposed in this invention, the confidence-layered guidance module is the core function of transforming the future situation information output from upstream into behavioral guidance instructions for multiple objectives and scenarios. The design of this module closely addresses practical maritime issues such as the complexity of port scheduling, the unreachability of maritime communication broadcasts, and the sensitivity of guidance risks. It proposes a guidance strategy framework based on dynamic layering of prediction confidence, thereby achieving a balance between information sharing and guidance intervention, effectively avoiding collective deviations or berth resource conflicts caused by misleading low-confidence instructions.

[0142] First, confidence modeling is required for the output of the prediction model. Due to limitations imposed by input data noise, dynamic edge structure, and the uncertainty of the target ship's own state during the modeling process, the reliability of predictions for different ships or different dimensions of the same ship varies significantly. Let the prediction result for the i-th ship at time t be:

[0143]

[0144] in, Let Δv be the future position offset vector. i For velocity change scalar, This represents the probability of berth occupancy. Let μ be the predicted distribution for the task category. An anomaly metric based on Mahalanobis distance is introduced to quantify the confidence level of each predicted value. Let μ be the mean vector of the outputs for this class in the training set. y The covariance matrix is ​​∑ y The overall confidence level of the current prediction for the i-th ship is denoted as:

[0145]

[0146] The confidence score s i ∈(0,1], the closer its value is to 1, the closer the prediction result is to the training distribution center and the more reliable it is, otherwise it means that the result is in the "unknown domain" or the range of data anomalies.

[0147] Based on the aforementioned confidence index, a hierarchical structure for the guidance strategy is constructed. This invention proposes three guidance levels, each corresponding to a different confidence interval:

[0148] When s i When the speed is ≥τH, the system generates direct control guidance commands, such as explicit speed adjustments, path corrections, or berth changes;

[0149] When τ L ≤s i <τ H At this time, the system generates suggested boot instructions, providing multiple alternative paths or suggested waiting times;

[0150] When s i <τ L At this time, the system only generates passive broadcast reminders, such as "It is recommended to reduce speed in high-density areas of the waterway" to avoid excessive intervention.

[0151] The above τ L τ H These are the low and high thresholds, for example, they can be 0.60 and 0.85 respectively. They can be dynamically set according to the risk level of port operations, or they can be automatically adjusted according to the model training results.

[0152] The process of generating guidance content requires a comprehensive consideration of multiple decision-making indicators, including the probability of future berth occupancy conflicts. Target path density With regard to the urgency of the current task

[0153] The probability of berth conflict is calculated as follows:

[0154]

[0155] in, sim represents the predicted berthing time for the i-th ship. T (·) indicates a time period overlap function; if the estimated berthing time difference of multiple ships at the same berth is less than the threshold, it is considered that there is a risk of conflict.

[0156] Path density The calculation of historical track clustering within a region is defined as follows:

[0157] in, For p i Let r be the set of track points in the spatial neighborhood of the center radius r, and δ be the length of the prediction time window;

[0158] It can be extracted directly based on task type weight or scheduling level field.

[0159] Ultimately, the decision weights are combined using the following aggregation function:

[0160]

[0161] Where ω1, ω2, and ω3 are normalized weighting coefficients, for high-confidence ships (i.e., s i ≥τ H And R i For individuals exceeding a given threshold, the system will issue detailed guidance instructions via NAVDAT structured communication broadcast frames, including recommended speeds, berth order adjustments, and port entry waiting strategies. The instruction content is encoded in a unified XML or ASN.1 pattern to support unified parsing by existing receivers.

[0162] To achieve the system's adaptive adjustment capability, a predictive execution feedback comparison mechanism can be further introduced. Let the actual behavior of the ship after execution be... The original predicted location was The error rate is then calculated as follows:

[0163]

[0164] Where ε is a very small positive number to prevent division by zero; when a ship is in K consecutive predictions ∈ i >∈ max If the behavior is deemed to deviate from the current distribution model, the online update mechanism of the confidence model will be automatically triggered.

[0165] At this point, the system retrieves the ship's most recent M prediction results. Calculate the local mean and covariance:

[0166]

[0167] Then, by weighting the fusion factor λ with the global model, it is updated as follows:

[0168]

[0169] Finally, the confidence level s is recalculated. i upd :

[0170]

[0171] Through the aforementioned feedback-driven mechanism, the system can adjust the cognitive model in a timely manner based on actual execution deviations, improve the dynamic adaptability of confidence assessment, truly achieve integrated closed-loop linkage of "prediction-guidance-feedback", give full play to the advantages of NAVDAT broadcast structure and high consistency, and ensure the effectiveness and robustness of the guidance strategy.

[0172] 4. Optimization of Ship Response Feedback

[0173] In this invention, the ship response feedback optimization module undertakes the functions of model refinement and guidance strategy evolution based on structured communication feedback. It is built upon the unique broadband structured broadcast and unified backhaul protocol capabilities of the NAVDAT system. Traditional AIS or VDES systems, due to low communication bandwidth, limited backhaul fields, or inconsistent feedback timing, cannot support high-density behavioral response acquisition across multiple targets and fields. This makes it difficult for behavior prediction systems to form structured attributions and dynamically adjust the deviation between actual ship actions and model predictions. This invention fully utilizes the structured frame field design and high-bandwidth link capabilities of the NAVDAT communication protocol. While the system distributes prediction and guidance information, it simultaneously broadcasts a unified feedback format and backhaul window to all target ships by embedding a "feedback response definition segment" field. This technically achieves standardized acquisition of multi-dimensional behavioral feedback, providing fundamental data support for feedback-driven model optimization and guidance scheduling.

[0174] Specifically, after the system completes the prediction and guidance of target vessels in a certain area, it defines a feedback protocol segment through a NAVDAT broadcast frame, explicitly requiring the vessel to report its execution results in a structured format within a specified response window [t, t+δ]. The feedback content includes multi-dimensional indicators such as trajectory deviation, speed execution curve, berth status label, guidance response intent, and task completion status, which together constitute a feedback behavior vector. This vector will be compared with the prediction guidance originally released by the system. Perform a structured comparison to generate a response bias vector:

[0175] And further calculate the standardized deviation rate index:

[0176] Where ||·||² represents the Euclidean norm, and ε > 0 is the smoothing regularization factor to avoid division by zero. The NAVDAT protocol supports feedback collection from hundreds of ships per cycle, enabling the system to quickly construct a structured response sample set.

[0177]

[0178] in This dataset serves as the environmental context feature vector for the ship at that time, including tidal state, meteorological parameters, port resource status, and scheduling level, to support normalized modeling under heterogeneous feedback. Based on this dataset, the system trains the feedback bias scoring function f. score Used to evaluate the expected response error under specific forecast and environment:

[0179]

[0180] This scoring function can be implemented based on kernel regression, gradient boosting trees, or neural network models to approximate the nonlinear response relationship between feedback error and input command.

[0181] After this scoring model is trained, it will be used to adjust the prediction broadcasting strategy for the next round, reducing the weights or correcting the scheduling penalty for instruction strategies with potentially large feedback biases. The specific adjustment mechanism is as follows: for any prediction guidance vector... The system calculates its deviation risk using a scoring function:

[0182] Based on the set feedback robustness expectation, the release command is modulated as follows:

[0183]

[0184] Where α∈[0,1] are modulation coefficients used to balance feedback risk and scheduling execution efficiency. This mechanism ensures the automatic compression of high-bias risk components in the strategy, improving the consistency and adaptability of global strategy execution.

[0185] Furthermore, the system can construct a response deviation space based on high-density structured feedback. Density clustering (such as DBSCAN) is used to identify behavioral response clusters with common deviation characteristics, and these clusters are assigned behavioral labels φ. k Forming a set of response anomaly tags Each anomalous cluster represents a subtype of navigation behavior that is systematically difficult to predict under the existing model structure, such as "waiting after short-distance crossing", "failed berth skidding", and "abnormal collision avoidance behavior". These cluster labels are structurally injected back into the predictive model training process to refine the boundary distribution, improve the ability to model minority behaviors, and influence the subsequent broadcast strategy release logic as dynamic scheduling constraint parameters.

[0186] In summary, this module, through the high-density, multi-field, and unified structured feedback channels supported by NAVDAT, constructs a system-level feedback-driven guidance strategy evolution mechanism, playing a crucial pivotal role in the "prediction-execution-feedback-optimization" closed loop. Its implementation relies not only on the feedback response itself but also on the irreplaceable communication capabilities provided by NAVDAT in terms of unified structure, feedback frequency, broadcast scheduling, and protocol flexibility. This allows for the construction of a multi-ship cooperative behavior modeling and strategy evolution platform that is difficult to establish under the traditional AIS / VDES framework.

[0187] Figure 3 This is a schematic diagram of the structure of a ship traffic intelligent prediction and cooperative guidance system based on NAVDAT communication according to an embodiment of the present invention. Figure 3 As shown, the system of the implementation scheme may include: a structured information parsing module 310, used to collect NAVDAT broadcast data and process the collected data to obtain a uniform format tensor. Where r is the length of the time window, and d * The system includes: a traffic state modeling and prediction module 320, used to model and predict traffic states based on the data processed by the structured information parsing module; a confidence-based hierarchical guidance module 330, used to transform the obtained future situation information into behavioral guidance instructions for multiple targets and scenarios; and a ship response feedback optimization module 340, used to optimize ship response feedback, including simultaneously broadcasting a unified feedback format and return window to all target ships by embedding feedback response definition field while issuing prediction and guidance information, thereby achieving standardized collection of multi-dimensional behavioral feedback and providing basic data support for feedback-driven model optimization and guidance scheduling.

[0188] The embodiments of the present invention have been described above by way of example, but the present invention is not limited to the embodiments described above. The basic idea of ​​the present invention lies in the above basic scheme. For those skilled in the art, designing various modified models, formulas, and parameters based on the teachings of the present invention does not require creative effort. Changes, modifications, substitutions, and variations made to the embodiments without departing from the principles and spirit of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication, characterized in that, include: 1) Collect NAVDAT broadcast data and process the collected data to obtain a uniformly formatted tensor. Where r is the time window length, d * This represents the number of effective feature dimensions after processing. 2) Based on the processed data from step 1), perform traffic condition modeling and prediction, including: 2.1) Construct a heterogeneous graph structure based on all target ships within the current time window. The node set V contains all individual ships that appear within the sliding window [t-r+1,t], and each node... Each node corresponds to a single ship; the edge set ε represents the spatial proximity, similar task relationships, or port scheduling correlations between ships; the attribute vector of each node is defined as the feature of its structured state at time t: , where d t (i) Let be the standardized structured vector of the i-th ship at time t; To quantify the boundary rights relationships between ships, two nodes v are defined. i With v j The edge weight is e ij Its calculation is based on the spatial distance between the two ships at time t, mission consistency, and environmental similarity, and is defined as: in , Let be the position coordinates of the i-th and j-th ships, respectively, and σ be the distance scale adjustment factor. This is a task type consistency indicator function, taking the value 1 or 0. This represents the cosine similarity based on the meteorological and tidal fields, where α, β, and γ are the weighting factors for the three types of relationships. 2.2) An attention-based graph neural network is used for node state updates and feature fusion. Information aggregation is achieved through the graph attention network to obtain the global embedding representation h of each ship. i (L) , 2.3) A multi-task prediction head is used to jointly predict the future state of each ship, and the output vector includes the position offset Δp. i Velocity change Δv i berth occupancy probability q i berth and the task transition probability distribution q i task The prediction formula is as follows: in W p With b p These are the prediction layer weights and biases, respectively. 3) Transform the future situation information obtained in step 2) into behavioral guidance instructions for multiple objectives and scenarios, including: 3.1) Evaluate the confidence level of the prediction output in step 2) to obtain the overall confidence level S of the current prediction for the i-th ship. i : 3.2) Based on the above confidence index S i Construct a hierarchical structure for the bootstrapping strategy: when s i ≥τ H When τ, the system generates direct control boot instructions; when τ L ≤s i <τ H When this happens, the system generates suggestive boot instructions, providing multiple alternative paths or suggested waiting times; when s i <τ L At that time, the system only generates passive broadcast alerts, where τ H τ L These are the high and low thresholds, respectively. 4) Optimize ship response feedback, including simultaneously broadcasting a unified feedback format and return window to all target ships by embedding feedback response definition fields while issuing prediction and guidance information, thereby achieving standardized collection of multi-dimensional behavioral feedback and providing basic data support for feedback-driven model optimization and guidance scheduling.

2. The intelligent prediction and collaborative guidance method for ship traffic based on NAVDAT communication according to claim 1, characterized in that, In step 1), the received NAVDAT broadcast data contains the following structured fields: ship dynamic information, environmental status information, and mission and scheduling information. The ship dynamic information includes geographic coordinates, speed, and heading, and the environmental status information includes weather codes. t and tidal level c t Task and scheduling information includes estimated berthing time T. t ETA Berth number b t Task type m t .

3. The method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to claim 2, characterized in that, In step 1), data processing includes: 1.1) Encode the field at time t into a unified structured feature vector: ， in, , indicating the ship's position; v t θt is the speed; θt is the heading angle; d is the total feature dimension. 1.2) Construct the input sequence based on the sliding time window mechanism to form a time-consistent dataset. Let the time window length be r, then the input window at time t is: ； 1.3) Perform missing data completion, including: Suppose that the nth field within the sliding window has a missing value at time point k. Its complete estimate is defined as: Where, μ n σ n These are the mean and standard deviation of field n, respectively. This is the offset adjustment factor. For smoothing terms, The original value of field n at time points other than k; 1.4) After completion, perform standardization processing on all numeric fields, including: Let μ be the mean of the g-th field in the completed structured feature vector within the window. g The standard deviation is σ g The standardized value is: ,in Let g be the original value of the g-th field at time t. After standardization, a data matrix with zero mean and unit variance is obtained: ; The completed and standardized sliding window sequence is denoted as W. t * ,Right now: ; Then let d * ≤d, we obtain a unified tensor: .

4. The intelligent prediction and collaborative guidance method for ship traffic based on NAVDAT communication according to claim 1, characterized in that, Step 2.2) includes: Update node v i The representation of the l-th layer is: ， in Let represent the set of neighbors connected to node i. Let l be the linear transformation matrix of the l-th layer. Indicates the activation function; Attention coefficient; The attention coefficient reflects the degree of information contribution of node j to node i, and is calculated as follows: in, Let be a learnable attention vector, and || denote the vector concatenation operation; After L layers of aggregation, a global embedded representation of each ship is obtained. This indicates that its timing status, the influence of neighboring vessels, and the scheduling environment have been integrated.

5. The method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to claim 1, characterized in that, Step 3.1) includes: Let the prediction result for the i-th ship at time t be: in, This is the future position offset vector. For velocity change scalar, This represents the probability of berth occupancy. The predicted distribution for task categories; An anomaly measure based on Mahalanobis distance quantifies the confidence level of each predicted value. Let the mean vector of the output be μ. y The covariance matrix is ​​Σ y Then the overall confidence level of the current prediction for the i-th ship is: Confidence score s i ∈(0,1], the closer the value is to 1, the more reliable the prediction result is.

6. The method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to claim 5, characterized in that, Step 3.2) includes combining the decision weights R i Provide guidance: Where ω1, ω2, ω3 are normalized weighting coefficients. For the probability of future berth occupancy conflicts, For the target path density, The urgency level of the current task; For high-confidence ships, i.e., s i ≥τ H And R i For vessels exceeding a given threshold, the system will issue detailed guidance instructions via NAVDAT structured communication broadcast frames, including recommended speed, berth sequence adjustments, and port entry waiting strategies; The probability of berth conflict is calculated as follows: , This represents the predicted berthing time for the i-th ship. Functions representing overlapping time periods; The calculation of historical track clustering within a region is defined as follows: ,in, For p i Let r be the set of track points in the spatial neighborhood of the center radius r, and δ be the length of the prediction time window; Extract directly based on task type weight or scheduling level field.

7. The method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to claim 6, characterized in that, Step 3) also includes introducing a prediction execution feedback comparison mechanism to recalculate the confidence level s. i upd , Let the actual behavior of the ship after execution be... The original predicted location was The error rate is then calculated as follows: ,in It must be a very small positive number to prevent division by zero; When a ship is predicted in K consecutive predictions If the behavior deviates from the current distribution model, the confidence level will be automatically recalculated, and the ship's most recent M prediction results will be retrieved. Calculate the local mean and covariance: Then, using the fusion factor λ and weighting, we obtain: Finally, the confidence level s is recalculated. i upd : 。 8. The method for intelligent prediction and collaborative guidance of ship traffic based on NAVDAT communication according to claim 1, characterized in that, Step 4) includes: 4.1) The feedback content of the ship within the specified response window [t, t+δ] uniformly constitutes the feedback behavior vector. This vector will be compared with the prediction guidance vector originally released by the system. Perform a structured comparison to generate a response bias vector: And further calculate the standardized deviation rate index: ， in Describes the Euclidean norm. >0 indicates a smoothing regularization factor; 4.2) Construct a structured response sample dataset based on step 4.1): , where c i (t) This is the feature vector of the ship's environmental context at that time; 4.3) Based on this dataset, train the feedback bias scoring function f. score Used to evaluate the expected response error under specific forecast and environment: ； 4.4) For any predicted guiding vector The risk of deviation is calculated using a scoring function: Based on the set feedback robustness expectation, the release command is modulated as follows: ,in ∈[0,1], which is the modulation coefficient.

9. The intelligent prediction and collaborative guidance method for ship traffic based on NAVDAT communication according to claim 8, characterized in that, It also includes 4.5), which constructs a response bias space based on structured feedback. Density clustering is used to identify behavioral response clusters with common deviation characteristics, and these clusters are then assigned behavioral labels. This forms a set of response exception labels A.

10. A ship traffic intelligent prediction and cooperative guidance system based on NAVDAT communication, characterized in that, include: The structured information parsing module is used to collect NAVDAT broadcast data and process the collected data to obtain a unified format tensor. Where r is the time window length, d * This represents the number of effective feature dimensions after processing. The traffic condition modeling and prediction module is used to perform traffic condition modeling and prediction based on the data processed by the structured information parsing module. The traffic condition modeling and prediction includes: 2.1) Construct a heterogeneous graph structure based on all target ships within the current time window. The node set V contains all individual ships that appear within the sliding window [t-r+1,t], and each node... Each node corresponds to a single ship; the edge set ε represents the spatial proximity, similar task relationships, or port scheduling correlations between ships; the attribute vector of each node is defined as the feature of its structured state at time t: , where d t (i) Let be the standardized structured vector of the i-th ship at time t; To quantify the boundary rights relationships between ships, two nodes v are defined. i With v j The edge weight is e ij Its calculation is based on the spatial distance between the two ships at time t, mission consistency, and environmental similarity, and is defined as: in , Let be the position coordinates of the i-th and j-th ships, respectively, and σ be the distance scale adjustment factor. This is a task type consistency indicator function, taking the value 1 or 0. This represents the cosine similarity based on the meteorological and tidal fields, where α, β, and γ are the weighting factors for the three types of relationships. 2.2) An attention-based graph neural network is used for node state updates and feature fusion. Information aggregation is achieved through the graph attention network to obtain the global embedding representation h of each ship. i (L) , 2.3) A multi-task prediction head is used to jointly predict the future state of each ship, and the output vector includes the position offset Δp. i Velocity change Δv i berth occupancy probability q i berth and the task transition probability distribution q i task The prediction formula is as follows: in W p With b p These are the prediction layer weights and biases, respectively. The confidence-based hierarchical guidance module is used to transform the obtained future situation information into behavioral guidance instructions for multiple targets and scenarios, including: 3.1) Evaluate the confidence level of the prediction output of the traffic state modeling and prediction module to obtain the overall confidence level S of the current prediction for the i-th ship. i : 3.2) Based on the above confidence index S i Construct a hierarchical structure for the bootstrapping strategy: when s i ≥τ H When τ, the system generates direct control boot instructions; when τ L ≤s i <τ H When this happens, the system generates suggestive boot instructions, providing multiple alternative paths or suggested waiting times; when s i <τ L At that time, the system only generates passive broadcast alerts, where τ H τ L These are the high and low thresholds, respectively. The ship response feedback optimization module is used to optimize ship response feedback. This includes simultaneously broadcasting a unified feedback format and return window to all target ships by embedding feedback response definition fields while issuing prediction and guidance information. This enables standardized collection of multi-dimensional behavioral feedback and provides basic data support for feedback-driven model optimization and guidance scheduling.

Images