A maritime traffic network mining method

Abstract

The application discloses a kind of offshore traffic network mining method, belong to offshore traffic field, comprising the following steps: S1, obtain standardized feature matrix, ship trajectory set and meteorological disturbance coefficient;S2, based on standardized feature matrix, AIS data, ship static parameter and wind farm data, fusion ship 3D safety area parameter and adaptive graph learning, construct 3D space constraint offshore traffic dynamic graph;S3, based on offshore traffic dynamic graph, fusion Patch-Transformer time series coding and information geometry multi-object coupling, obtain global spatiotemporal correlation matrix and key correlation path;S4, based on the feedback of key node and path, global spatiotemporal correlation matrix is corrected by iterative optimization.Using the offshore traffic network mining method described above, global spatiotemporal correlation can be accurately captured and key correlation path can be efficiently identified, improving the dynamic adaptability of network mining and decision support value.

Description

Technical Field

[0001] This invention relates to the field of maritime transportation technology, and in particular to a method for mining maritime transportation networks. Background Technology

[0002] With the growth of maritime trade, the large-scale construction of offshore wind farms, and the expansion of marine economic activities such as coastal tourism, maritime traffic scenarios are becoming increasingly complex. Maritime traffic network mining, as a core technology for intelligent maritime management, aims to extract key waterways, identify risk nodes, and optimize traffic flow configuration by analyzing the correlation between ship trajectories, facility distribution, and environmental parameters, providing decision support for collision warnings, route planning, and emergency response. However, existing maritime traffic network modeling technologies are mostly based on "two-dimensional static geospatial association" to construct network structures, failing to fully adapt to the three-dimensional spatial characteristics and dynamic changes of maritime traffic, resulting in a disconnect between network models and actual navigation safety requirements. Summary of the Invention

[0003] The purpose of this invention is to provide a method for mining maritime traffic networks to solve the above-mentioned technical problems.

[0004] To achieve the above objectives, the present invention provides a method for mining marine traffic networks, comprising the following steps:

[0005] S1. Collect multi-source heterogeneous data consisting of AIS data, ship static parameters, meteorological parameters and wind farm data, and preprocess it to obtain a standardized feature matrix, ship trajectory set and meteorological interference coefficient.

[0006] S2. Based on standardized feature matrices, AIS data, ship static parameters and wind farm data, the ship 3D safety domain parameters and adaptive graph learning are integrated to construct a 3D spatially constrained dynamic map of maritime traffic that changes with ship status and weather dynamics.

[0007] S3. Based on the dynamic map of maritime traffic, Patch-Transformer temporal coding and information geometric multi-objective coupling are integrated to obtain the global spatiotemporal correlation matrix and key correlation paths;

[0008] S4. Based on the feedback from key nodes and paths, the global spatiotemporal correlation matrix is ​​iteratively optimized and corrected.

[0009] Therefore, the above-mentioned method for mining marine traffic networks has the following beneficial effects:

[0010] Improve the accuracy of multi-source heterogeneous data processing: AIS data, ship static parameters, meteorological parameters and wind farm data can be collected simultaneously and standardized and integrated. By constructing PS manifold features and using Bregman divergence to calculate the geometric mean of normal clutter, AIS abnormal data can be accurately identified and cleaned to obtain a high-quality set of ship trajectories and a standardized feature matrix, laying a reliable data foundation for subsequent network mining.

[0011] Achieving dynamic adaptive modeling of maritime traffic networks: Breaking through the limitations of traditional two-dimensional static networks, this method integrates ship nodes, wind turbine fixed nodes, and key waterway nodes to form a node set. It then integrates geographical distance (calculated based on the Havesing formula of the Earth's average radius), 3D safety domain spatial overlap, and ship maneuver distance to generate an initial edge weight matrix. Finally, it combines adaptive graph learning to mine potential relationships between nodes and constructs a 3D spatially constrained dynamic maritime traffic graph that changes dynamically with ship status and weather conditions, making it more consistent with the characteristics of actual maritime traffic scenarios.

[0012] Improving the accuracy of global spatiotemporal correlation capture: By extracting and embedding the patch features of the maritime traffic dynamic map, the dynamic coupling relationship of multiple ships is quantified by combining the PS manifold features within the patch with the Regman divergence. The self-attention mechanism of Patch-Transformer temporal coding is used to capture long-range spatiotemporal correlations across patches. At the same time, the coupling correlation is used as an attention weight correction term to optimize the correlation weight of the multi-ship coupling region. Finally, the global spatiotemporal correlation matrix is ​​corrected through iterative optimization, which effectively improves the accuracy and comprehensiveness of spatiotemporal correlation identification.

[0013] To ensure the effectiveness of critical associated path identification: First, candidate associated edges are screened based on the global spatiotemporal association strength threshold. Then, high-risk candidate edges are filtered through a comprehensive risk model (integrating forward risk, starboard risk, draft risk, and airspace risk). A depth-first search is used to traverse the screened candidate edges, and a comprehensive path score (combining association strength and risk correction coefficient) is calculated. Paths with a score greater than 0.85 are selected as critical associated paths to ensure that the identification results have both high correlation and navigation safety.

[0014] Enhance the adaptability to actual maritime scenarios: Integrate wind turbine node attributes (longitude, latitude, and lower clearance) and meteorological interference coefficients into network modeling to fully consider the impact of actual environmental factors such as offshore wind farms and complex weather on the transportation network; The key correlation paths and global spatiotemporal correlation matrix discovered can directly provide accurate decision support for intelligent maritime management work such as collision warning, route planning, and emergency response, adapting to diverse maritime traffic management needs.

[0015] In summary, by integrating multi-source heterogeneous data from AIS, ship static parameters, meteorology, and wind farms, and combining ship 3D safety domain parameters, adaptive graph learning, and Patch-Transformer temporal coding technology, a 3D spatially constrained dynamic map of maritime traffic that dynamically changes with ship status and weather can be constructed. This can accurately capture global spatiotemporal correlations and efficiently identify key correlation paths, thereby improving the dynamic adaptability and decision support value of network mining.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for mining a marine traffic network according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0019] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0020] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] like Figure 1 As shown, a method for mining a maritime traffic network includes the following steps:

[0022] S1. Collect multi-source heterogeneous data consisting of AIS data, ship static parameters, meteorological parameters and wind farm data, and preprocess it to obtain a standardized feature matrix, ship trajectory set and meteorological interference coefficient.

[0023] Step S1 specifically includes the following steps:

[0024] S11. Synchronously acquire AIS data Ship static parameters Meteorological parameters Wind farm data Obtain multi-source heterogeneous data ,in, , Representing ships Longitude, latitude, ground speed, ground heading, and timestamp; , Representing ships The ship's length, maximum draft, air draft, and FSAH-FAS stopping distance; , Representing timestamps The corresponding wave height, wind speed, visibility, and ocean current speed; , These represent the wind turbines in the wind farm. Longitude and latitude;

[0025] S12, Pretreatment;

[0026] S121. Perform anomaly detection and cleaning on the AIS data in the multi-source heterogeneous data to obtain the set of ship trajectories after anomaly cleaning. ;

[0027] S122. Extract AIS dynamic features, ship 3D safety domain basic features, and meteorological interference features from the cleaned AIS data.

[0028] S123. The AIS dynamic features, ship 3D safety domain basic features, AIS dynamic features and meteorological interference features are fused to obtain the fused feature vector. ;

[0029] S124. For the fused feature vector Spatiotemporal standardization is performed to obtain the standardized feature matrix. .

[0030] Step S121 specifically includes the following steps:

[0031] S1211. Constructing PS manifold features: [This section appears to be incomplete and requires further context.] of Modeled as points on a PS manifold , Indicates the number of pulses;

[0032] S1212. Calculate the geometric mean of normal clutter on the PS manifold. :

[0033] ;

[0034] In the formula, Indicates the number of reference samples; Bregman divergence function, and , Represents the trace of a matrix. Represents the identity matrix. The natural logarithm of the determinant of a matrix; Indicates the first Feature matrix of one reference sample; Represent the set of points on the PS manifold;

[0035] S1213, Anomaly Detection and Cleaning: If If the condition is met, the AIS data point will be removed; otherwise, it will be retained. This indicates that a threshold value has been set.

[0036] The dynamic characteristics of AIS in S122 are as follows:

[0037] ;

[0038] ;

[0039] In the formula, and Representing ships exist The change in velocity relative to the ground and the change in heading relative to the ground at any given moment; and Separate ships exist At any given moment, the velocity relative to the ground and the heading relative to the ground;

[0040] The basic features of a ship's 3D safety domain include the initial domain depth. Initial domain height and initial forward domain length ;

[0041] The characteristics of meteorological disturbances are as follows:

[0042] ;

[0043] In the formula, Indicates the meteorological interference coefficient; and Representing timestamps The average wave height and average wind speed below;

[0044] In step S123, the obtained fused feature vector ,in, Representing ships The relative longitude and latitude, and , and Representing ships The absolute longitude and latitude.

[0045] S2. Based on standardized feature matrices, AIS data, ship static parameters and wind farm data, the ship 3D safety domain parameters and adaptive graph learning are integrated to construct a 3D spatially constrained dynamic map of maritime traffic that changes with ship status and weather dynamics.

[0046] Step S2 specifically includes the following steps:

[0047] S21, Integrating Ship Nodes Wind turbine nodes and key nodes of the waterway , obtain the node set Among them, ship nodes , Indicates a ship The node corresponding to the timestamp, and its attributes include , Indicates a ship timestamp The standardized feature vector, Indicates a ship Maximum draft; wind turbine nodes , Indicates wind turbine The corresponding fixed node, and its attributes include and , Indicates wind turbine Clearance below; key nodes of the waterway Including channel junctions and turning points;

[0048] S22, Integrated Geographical Distance Overlap with 3D security domain space Calculate the initial edge weight matrix :

[0049] ;

[0050] in,

[0051] ;

[0052] ;

[0053] ;

[0054] ;

[0055] In the formula, Represents a node With nodes Initial edge weights between them; , and All represent weighting coefficients; Indicates the average radius of the Earth; and Representing nodes respectively and nodes Latitude; Represents a node and nodes The difference in latitude; Represents a node and nodes The difference in longitude; Represents a node and nodes 3D security domain overlap volume; and Representing nodes respectively and nodes 3D security domain volume; and Representing nodes respectively and nodes The FSAH-FAS stopping distance refers to the total distance a vessel travels from switching from "full speed forward" to "full speed reverse" until it comes to a complete stop. This is a constant term, in this embodiment Take 10 -3 ;

[0056] S23. Utilize adaptive graph learning to uncover potential relationships between nodes and fuse the initial edge weight matrix. The dynamic adjacency matrix is ​​obtained. :

[0057] ;

[0058] in,

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] In the formula, Indicates the fusion coefficient; express The adaptive adjacency matrix at time step; Indicates the row direction (softmax); Represents the normalized node and nodes The potential strength of the association; Indicates the softmax temperature coefficient; Represents a node and nodes The potential strength of the association; Indicates transpose; and These represent two types of projective embedding vectors for node attributes; Represents the node attribute matrix; and Both represent projected weights; and Both represent bias terms;

[0065] S23. Filter risk-related edges using the RN risk index and perform sparsification to obtain the maritime traffic dynamics graph. , Representing time respectively The set of nodes, the set of edges, and the associated weights. Represents the nodes in the sparse adjacency matrix With nodes The association weight.

[0066] S3. Based on the dynamic map of maritime traffic, Patch-Transformer temporal coding and information geometric multi-objective coupling are integrated to obtain the global spatiotemporal correlation matrix and key correlation paths;

[0067] Step S3 specifically includes the following steps:

[0068] S31. Extract maritime traffic dynamic map Patch features in :

[0069] ;

[0070] In the formula, This represents the matrix flattening operation; Indicates feature concatenation operation; express The sparse adjacency matrix at time step; express The node-normalized feature matrix at time step;

[0071] S32, Project onto the embedding dimension and add sinusoidal position encoding:

[0072] ;

[0073] In the formula, Indicates an embedded sequence; Denotes the Patch feature projection weight matrix, and , and These represent the dimensions of the patch after it is flattened and the dimensions of the patch embedded, respectively. This represents the projection bias term of the Patch feature; Indicates the first The sinusoidal positional encoding vector of each patch;

[0074] S33. Based on the PS manifold characteristics within the Patch and the Regemann divergence metric, the dynamic coupling relationship of multiple ships within the Patch time series is quantified to obtain the Patch-level coupling correlation matrix. And a list of strongly coupled ship pairs;

[0075] S34. Combining Patch-Transformer temporal coding, Patch-level coupling correlation matrix, and strongly coupled ship pair list, long-range spatiotemporal correlations across Patch are captured through self-attention. At the same time, coupling correlation is used as an attention weight correction term to correct the correlation weights of multi-ship coupling regions, and a global spatiotemporal correlation matrix is ​​output.

[0076] Step S34 specifically includes the following steps:

[0077] S341. Embed the Patch into the sequence. Patch-level coupling correlation matrix Flattening and concatenating features yields the Transformer input sequence. ;

[0078] S342. Calculate the Transformer input sequence. Self-attention weights:

[0079] ;

[0080] In the formula, Patch With Patch Self-attention weights between them; Patch The query vector; Patch The key vector; Patch exist Key vectors under each attention head; Indicates the attention head dimension; Indicates the number of patches;

[0081] S343. Couple the Patch-level correlation matrix Integration The corrected self-attention weights are obtained as follows:

[0082] ;

[0083] In the formula, This indicates the corrected patch. With Patch Self-attention weights between them; Indicates the coupling correction coefficient; and They represent Patch Strongly coupled ship pairs and Patch A set of strongly coupled ship pairs;

[0084] S344. Obtain the global spatiotemporal correlation matrix through linear projection. :

[0085] ;

[0086] In the formula, Represents a linear projection function; Indicates Transformer encoder; Represents the input sequence of the Transformer; This represents the corrected self-attention weights.

[0087] Step S34 is followed by S35, extracting key association paths:

[0088] S351, Filtering candidate related edges , Represents a node With nodes The strength of the global spatiotemporal correlation between them Indicates the threshold of association strength;

[0089] S352, Utilizing candidate related edges Combined risks at both ends of the node Filtering candidate edges Among them, comprehensive risks The expression is as follows:

[0090] ;

[0091] In the formula, , , and All represent weighting coefficients; , , and These respectively represent forward risk, starboard risk, draft risk, and airspace risk;

[0092] S353. Use depth-first search to traverse the filtered candidate edges, extract connected paths with ≥3 nodes, and calculate the comprehensive path score. :

[0093] ;

[0094] In the formula, Indicates the number of candidate edges contained in the path;

[0095] S354, Filtering Obtain the key associated path. .

[0096] S4. Based on the feedback from key nodes and paths, the global spatiotemporal correlation matrix is ​​iteratively optimized and corrected.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for excavating a marine traffic network, characterized in that: Includes the following steps: S1. Collect multi-source heterogeneous data consisting of AIS data, ship static parameters, meteorological parameters and wind farm data, and preprocess it to obtain a standardized feature matrix, ship trajectory set and meteorological interference coefficient. S2. Based on standardized feature matrices, AIS data, ship static parameters and wind farm data, the ship 3D safety domain parameters and adaptive graph learning are integrated to construct a 3D spatially constrained dynamic map of maritime traffic that changes with ship status and weather dynamics. S3. Based on the dynamic map of maritime traffic, Patch-Transformer temporal coding and information geometric multi-objective coupling are integrated to obtain the global spatiotemporal correlation matrix and key correlation paths; S4. Based on the feedback from key nodes and paths, the global spatiotemporal correlation matrix is ​​iteratively optimized and corrected. Step S3 specifically includes the following steps: S31. Extract maritime traffic dynamic map Patch features in : ； In the formula, This represents the matrix flattening operation; Indicates feature concatenation operation; express The sparse adjacency matrix at time step; express The node-normalized feature matrix at time step; S32, Project onto the embedding dimension and add sinusoidal position encoding: ； In the formula, Represents an embedded sequence; Denotes the Patch feature projection weight matrix, and , and These represent the dimensions of the patch after it is flattened and the dimensions of the patch embedded, respectively. This represents the projection bias term of the Patch feature; Indicates the first The sinusoidal positional encoding vector of each patch; S33. Based on the PS manifold characteristics within the Patch and the Regemann divergence metric, the dynamic coupling relationship of multiple ships within the Patch time series is quantified to obtain the Patch-level coupling correlation matrix. And a list of strongly coupled ship pairs; S34. Combining Patch-Transformer temporal coding, Patch-level coupling correlation matrix, and strongly coupled ship pair list, long-range spatiotemporal correlations across Patch are captured through self-attention. At the same time, coupling correlation is used as an attention weight correction term to correct the correlation weights of multi-ship coupling regions, and a global spatiotemporal correlation matrix is ​​output.

2. The method for excavating a marine traffic network according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Synchronously acquire AIS data Ship static parameters Meteorological parameters Wind farm data Obtain multi-source heterogeneous data ,in, , Representing ships Longitude, latitude, ground speed, ground heading, and timestamp; , Representing ships The ship's length, maximum draft, air draft, and FSAH-FAS stopping distance; , Representing timestamps The corresponding wave height, wind speed, visibility, and ocean current speed; , These represent the wind turbines in the wind farm. Longitude and latitude; S12, Pretreatment; S121. Perform anomaly detection and cleaning on the AIS data in the multi-source heterogeneous data to obtain the set of ship trajectories after anomaly cleaning. ; S122. Extract AIS dynamic features, ship 3D safety domain basic features, and meteorological interference features from the cleaned AIS data. S123. The AIS dynamic features, ship 3D safety domain basic features, AIS dynamic features and meteorological interference features are fused to obtain the fused feature vector. ; S124. For the fused feature vector Spatiotemporal standardization is performed to obtain the standardized feature matrix. .

3. The method for excavating a maritime traffic network according to claim 2, characterized in that: Step S121 specifically includes the following steps: S1211. Constructing PS manifold features: [This section appears to be incomplete and requires further context.] of Modeled as points on a PS manifold , Indicates the number of pulses; S1212. Calculate the geometric mean of normal clutter on the PS manifold. : ； In the formula, Indicates the number of reference samples; Bregman divergence function, and , Represents the trace of a matrix. Represents the identity matrix. The natural logarithm of the determinant of a matrix; Indicates the first Feature matrix of one reference sample; Represent the set of points on the PS manifold; S1213, Anomaly Detection and Cleaning: If If the condition is met, the AIS data point will be removed; otherwise, it will be retained. This indicates that a threshold value has been set.

4. The method for excavating a maritime traffic network according to claim 2, characterized in that: The dynamic characteristics of AIS in S122 are as follows: ； ； In the formula, and Representing ships exist The change in velocity relative to the ground and the change in heading relative to the ground at any given moment; and Separate ships exist At any given moment, the velocity relative to the ground and the heading relative to the ground; The basic features of a ship's 3D safety domain include the initial domain depth. Initial domain height and initial forward domain length ; The characteristics of meteorological disturbances are as follows: ； In the formula, Indicates the meteorological interference coefficient; and Representing timestamps The average wave height and average wind speed below; In step S123, the obtained fused feature vector ,in, Representing ships The relative longitude and latitude, and , and Representing ships The absolute longitude and latitude.

5. The method for excavating a maritime traffic network according to claim 2, characterized in that: Step S2 specifically includes the following steps: S21, Integrating Ship Nodes Wind turbine nodes and key nodes of the waterway , obtain the node set Among them, ship nodes , Indicates a ship The node corresponding to the timestamp, and its attributes include , Indicates a ship timestamp The standardized feature vector, Indicates a ship Maximum draft; wind turbine nodes , Indicates wind turbine The corresponding fixed node, and its attributes include and , Indicates wind turbine Clearance below; key nodes of the waterway Including channel junctions and turning points; S22, Integrated Geographical Distance Overlap with 3D security domain space Calculate the initial edge weight matrix : ； in, ； ； ； ； In the formula, Represents a node With nodes Initial edge weights between them; , and All represent weighting coefficients; Indicates the average radius of the Earth; and Representing nodes respectively and nodes Latitude; Represents a node and nodes The difference in latitude; Represents a node and nodes The difference in longitude; Represents a node and nodes 3D security domain overlap volume; and Representing nodes respectively and nodes 3D security domain volume; and Representing nodes respectively and nodes FSAH-FAS stopping distance; For constant terms; S23. Utilize adaptive graph learning to uncover potential relationships between nodes and fuse the initial edge weight matrix. The dynamic adjacency matrix is ​​obtained. : ； in, ； ； ； ； ； In the formula, Indicates the fusion coefficient; express The adaptive adjacency matrix at time step; Indicates the row direction (softmax); Represents the normalized node and nodes The potential strength of the association; Indicates the softmax temperature coefficient; Represents a node and nodes The potential strength of the association; Indicates transpose; and These represent two types of projective embedding vectors for node attributes; Represents the node attribute matrix; and Both represent projected weights; and Both represent bias terms; S23. Filter risk-related edges using the RN risk index and perform sparsification to obtain the maritime traffic dynamics graph. , Representing time respectively The set of nodes, the set of edges, and the associated weights. Represents the nodes in the sparse adjacency matrix With nodes The association weight.

6. The method for excavating a marine traffic network according to claim 5, characterized in that: Step S34 specifically includes the following steps: S341. Embed the Patch into the sequence. Patch-level coupling correlation matrix Flattening and concatenating features yields the Transformer input sequence. ; S342. Calculate the Transformer input sequence. Self-attention weights: ； In the formula, Patch With Patch Self-attention weights between them; Patch The query vector; Patch The key vector; Patch exist Key vectors under each attention head; Indicates the attention head dimension; Indicates the number of patches; S343. Couple the Patch-level correlation matrix Integration The corrected self-attention weights are obtained as follows: ； In the formula, This indicates the corrected patch. With Patch Self-attention weights between them; Indicates the coupling correction coefficient; and They represent Patch Strongly coupled ship pairs and Patch A set of strongly coupled ship pairs; S344. Obtain the global spatiotemporal correlation matrix through linear projection. : ； In the formula, Represents a linear projection function; Indicates Transformer encoder; Represents the input sequence of the Transformer; This represents the corrected self-attention weights.

7. The method for excavating a marine traffic network according to claim 6, characterized in that: Step S34 is followed by S35, extracting key association paths: S351, Filtering candidate related edges , Represents a node With nodes The strength of the global spatiotemporal correlation between them Indicates the threshold of association strength; S352, Utilizing candidate related edges Comprehensive risks at both ends of the node Filtering candidate edges Among them, comprehensive risks The expression is as follows: ； In the formula, , , and All represent weighting coefficients; , , and These respectively represent forward risk, starboard risk, draft risk, and airspace risk; S353. Use depth-first search to traverse the filtered candidate edges, extract connected paths with ≥3 nodes, and calculate the comprehensive path score. : ； In the formula, Indicates the number of candidate edges contained in the path; S354, Filtering Obtain the key associated path. .

Images