Search

CN-121255792-B - Sea traffic network mining method

CN121255792BCN 121255792 BCN121255792 BCN 121255792BCN-121255792-B

Abstract

The invention discloses an offshore traffic network mining method, which belongs to the field of offshore traffic and comprises the following steps of S1, obtaining a standardized feature matrix, a ship track set and meteorological interference coefficients, S2, based on the standardized feature matrix, AIS data, ship static parameters and wind field data, fusing ship 3D safety domain parameters and self-adaptive graph learning, constructing an offshore traffic dynamic graph with 3D space constraint, S3, based on the offshore traffic dynamic graph, fusing Patch-transform time sequence coding and information geometric multi-target coupling, obtaining a global space-time correlation matrix and a key correlation path, and S4, based on feedback of key nodes and paths, correcting the global space-time correlation matrix through iterative optimization. By adopting the method for mining the offshore traffic network, the global time-space correlation can be accurately captured, the key correlation path can be efficiently identified, and the dynamic adaptability and decision support value of network mining are improved.

Inventors

  • CHANG YUGANG
  • JI JINGWEI
  • MAO WEI
  • Shen Gaohan
  • WANG CHUMING

Assignees

  • 中华人民共和国洋山港海事局

Dates

Publication Date
20260508
Application Date
20251028

Claims (7)

  1. 1. The method for excavating the offshore traffic network is characterized by comprising the following steps of: S1, acquiring multi-source heterogeneous data consisting of AIS data, ship static parameters, meteorological parameters and wind farm data, and preprocessing the multi-source heterogeneous data to obtain a standardized feature matrix, a ship track set and meteorological interference coefficients; S2, based on a standardized feature matrix, AIS data, ship static parameters and wind farm data, fusing ship 3D safety domain parameters and self-adaptive map learning, and constructing a marine traffic dynamic map constrained by a 3D space along with ship states and weather dynamic changes; S3, based on an offshore traffic dynamic diagram, integrating Patch-transform time sequence coding and information geometric multi-target coupling to obtain a global space-time correlation matrix and a key correlation path; s4, correcting the global space-time correlation matrix through iterative optimization based on feedback of the key nodes and the paths; The step S3 specifically comprises the following steps: S31, extracting an offshore traffic dynamic map Patch feature in (C) : ; In the formula, Representing a matrix flattening operation; Representing feature stitching operations; Representation of A sparse adjacency matrix of time; Representation of The node standardization characteristic matrix at moment; S32、 Projection to the embedding dimension and adding sinusoidal position coding: ; in the formula, Representing an embedded sequence; Representing a Patch feature projection weight matrix, an , And Representing the dimension of the Patch after flattening and the Patch embedding dimension respectively; representing a Patch feature projection bias term; Represent the first Sinusoidal position coding vectors for the Patch; S33, quantifying a dynamic coupling relation of multiple ships in a Patch time sequence based on PS manifold features and Lei Geman divergences in the Patch to obtain a Patch-level coupling correlation matrix A list of strongly coupled vessel pairs; S34, combining the Patch-transducer time sequence coding, the Patch-level coupling incidence matrix and the strong coupling ship pair list, performing self-attention capturing cross-Patch length Cheng Shikong association, and simultaneously using the coupling incidence as an attention weight correction term to correct the incidence weights of multiple ship coupling areas and outputting a global space-time incidence matrix.
  2. 2. The method for mining the marine transportation network according to claim 1, wherein the step S1 comprises the following steps: s11, synchronously collecting AIS data Static parameters of ship Meteorological parameters Wind farm data Obtaining multi-source heterogeneous data , wherein, , Respectively represent ships Longitude, latitude, speed to ground, heading to ground, and time stamp; , Respectively represent ships Is a long, maximum draft, air draft, FSAH-FAS stopping distance; , Respectively represent time stamps Corresponding wave height, wind speed, visibility and ocean current speed; , Respectively represent fans in wind power plant Longitude and latitude of (a); S12, preprocessing; s121, detecting and cleaning abnormal data of AIS data in the multi-source heterogeneous data to obtain a ship track set after abnormal cleaning ; S122, extracting AIS dynamic characteristics, ship 3D safety domain basic characteristics and meteorological interference characteristics in AIS data subjected to data cleaning; s123, fusing the AIS dynamic characteristics, the ship 3D safety domain basic characteristics, the AIS dynamic characteristics and the weather interference characteristics to obtain fusion characteristic vectors ; S124, fusing feature vectors Performing space-time standardization processing to obtain a standardized feature matrix 。
  3. 3. The method for mining the marine transportation network according to claim 2, wherein the step S121 comprises the following steps: S1211, constructing PS manifold feature, namely, ship A kind of electronic device Modeled as points on a PS manifold , The number of pulses is represented; s1212, calculating the geometric mean of the normal clutter on the PS manifold : ; In the formula, Representing the number of reference samples; a Bragg divergence function, an , The trace of the matrix is indicated, Representing the identity matrix of the cell, Representing the natural logarithm of the matrix determinant; Represent the first Feature matrices of the reference samples; Representing a set of points on the PS manifold; s1213, abnormality determination and cleaning if Rejecting the AIS data point, otherwise, keeping; indicating that the decision threshold is set.
  4. 4. The method for mining an offshore transportation network according to claim 2, wherein the AIS dynamic characteristics in S122 are as follows: ; ; in the formula, And Respectively represent ships At the position of The change of the speed and the change of the heading to the earth at the moment; And Separate vessels At the position of The speed and heading to the earth at the moment; the ship 3D security domain base features include an initial domain depth Initial domain height And an initial forward domain length ; The weather disturbance is characterized as follows: ; in the formula, Representing a meteorological interference coefficient; And Respectively represent time stamps Average wave height and average wind speed under; In step S123, the obtained fusion feature vector , wherein, Respectively represent ships And (2) relative longitude and latitude, and , And Respectively represent ships Absolute longitude and latitude of (a).
  5. 5. The method for mining the marine transportation network according to claim 2, wherein the step S2 comprises the following steps: S21, integrated ship node Fan node And channel key node Obtaining a node set Wherein, the ship node , Representing a ship The node corresponds to the position of the time stamp, and the attribute comprises , Representing a ship At the time stamp Is used for the normalization feature vector of (a), Representing a ship Maximum draft of (d) fan node , Indicating fan Corresponding fixed nodes, and their attributes include And , Indicating fan Lower clearance of (2) channel key node The system comprises a channel intersection point and a steering point; S22, fusing geographic distance And 3D security domain spatial overlap Calculating an initial edge weight matrix : ; Wherein, the ; ; ; ; In the formula, Representing nodes And node Initial edge weights in between; 、 And All represent weight coefficients; Represents the average radius of the earth; And Representing nodes respectively Sum node Latitude of (a); Representing nodes Sum node Latitude difference of (2); Representing nodes Sum node Longitude differences of (2); Representing nodes Sum node A 3D security domain overlap volume of (2); And Representing nodes respectively Sum node 3D security domain volume of (2); And Representing nodes respectively Sum node FSAH-FAS stopping distance; Is a constant term; S23, learning and mining potential relations among nodes by utilizing the self-adaptive graph, and fusing an initial edge weight matrix Obtaining dynamic adjacency matrix : ; Wherein, the ; ; ; ; ; In the formula, Representing the fusion coefficient; Representation of An adaptive adjacency matrix of time; representing a row direction softmax; Representing normalized nodes Sum node Is a potential association strength of (a); Representing a softmax temperature coefficient; Representing nodes Sum node Is a potential association strength of (a); representing a transpose; And Two projection embedded vectors respectively representing node attributes; representing a node attribute matrix; And All represent projection weights; And All represent bias terms; s23, filtering risk associated edges by using the RN risk indexes, and performing sparsification treatment to obtain an offshore traffic dynamic map , Respectively represent time of day Node sets, edge sets and associated weights, Representing nodes in a sparse adjacency matrix And node Is used to determine the associated weight of the (c).
  6. 6. The method for mining an offshore transportation network according to claim 5, wherein the step S34 comprises the steps of: s341, embedding Patch into sequence Coupling incidence matrix with Patch stage Flattening feature stitching to obtain a transducer input sequence ; S342, calculating a transducer input sequence Is a self-attention weight of (2): ; in the formula, Representing Patch And Patch A self-attention weight in between; Representing Patch Is a query vector of (1); Representing Patch Is a key vector of (a); Representing Patch At the position of Key vectors under the individual attention heads; Representing the attention head dimension; representing the number of Patches; s343, coupling the Patch level with the incidence matrix Is integrated into Obtaining corrected self-attention weight: ; in the formula, Representing modified Patch And Patch A self-attention weight in between; representing the coupling correction coefficient; And Respectively represent Patch Is a strong coupling ship pair set and Patch Is a set of strongly coupled vessel pairs; s344, obtaining a global space-time correlation matrix through linear projection : ; In the formula, Representing a linear projection function; representing a transducer encoder; representing a transducer input sequence; Representing the corrected self-attention weight.
  7. 7. The method for mining an offshore transportation network according to claim 6, wherein the step S34 further comprises the step S35 of extracting key associated paths: s351, screening candidate associated edges , Representing nodes And node The strength of the global spatiotemporal association between, Representing an association strength threshold; S352, utilizing candidate associated edges Comprehensive risk of two-end nodes Filtering candidate edges Wherein the risk is integrated The expression is as follows: ; in the formula, 、 、 And All represent weight coefficients; 、 、 And Respectively representing a forward risk, a starboard risk, a draft risk and a clearance risk; S353, traversing the filtered candidate edges by adopting depth-first search, extracting the connected paths with node number more than or equal to 3, and calculating the path comprehensive scores : ; In the formula, Representing the number of candidate edges contained in the path; s354, screening Obtain a critical associated path 。

Description

Sea traffic network mining method Technical Field The invention relates to the technical field of offshore traffic, in particular to an offshore traffic network mining method. Background With the increase of maritime traffic trade volume, the large-scale construction of offshore wind power stations, the expansion of maritime economic activities such as coastal travel and the like, the maritime traffic scene is increasingly complex. The core aim of the maritime traffic network mining is to extract a key channel, identify risk nodes and optimize traffic flow configuration by analyzing the association relation between ship track, facility distribution and environmental parameters, and provide decision support for collision early warning, route planning and emergency response. However, the existing marine traffic network modeling technology is mostly based on the 'geospatial two-dimensional static correlation' to construct a network structure, and the network model is disjointed from the actual navigation safety requirement due to the fact that the three-dimensional spatial characteristics and the dynamic change rule of the marine traffic are not fully adapted. Disclosure of Invention The invention aims to provide an offshore traffic network mining method which solves the technical problems. In order to achieve the above object, the present invention provides an offshore transportation network mining method, comprising the steps of: S1, acquiring multi-source heterogeneous data consisting of AIS data, ship static parameters, meteorological parameters and wind farm data, and preprocessing the multi-source heterogeneous data to obtain a standardized feature matrix, a ship track set and meteorological interference coefficients; S2, based on a standardized feature matrix, AIS data, ship static parameters and wind farm data, fusing ship 3D safety domain parameters and self-adaptive map learning, and constructing a marine traffic dynamic map constrained by a 3D space along with ship states and weather dynamic changes; S3, based on an offshore traffic dynamic diagram, integrating Patch-transform time sequence coding and information geometric multi-target coupling to obtain a global space-time correlation matrix and a key correlation path; s4, correcting the global space-time correlation matrix through iterative optimization based on feedback of the key nodes and the paths. Therefore, the invention adopts the method for excavating the offshore traffic network, and has the beneficial effects that: The multi-source heterogeneous data processing accuracy is improved, namely AIS data, ship static parameters, meteorological parameters and wind farm data can be synchronously acquired and integrated in a standardized manner, and a reliable data base is laid for subsequent network mining by constructing PS manifold features and calculating normal clutter geometric mean values by utilizing Bragg divergence, accurately judging and cleaning AIS abnormal data to obtain a high-quality ship track set and a standardized feature matrix; Breaking through the limitation of the traditional two-dimensional static network, integrating ship nodes, fan fixed nodes and channel key nodes to form a node set, fusing geographic distance (Ha Fuxin formula calculation based on the average radius of the earth), 3D safety domain space overlapping degree and ship maneuver distance to generate an initial side weight matrix, and combining an adaptive graph to learn potential association of excavation nodes to construct a 3D space constraint sea traffic dynamic graph which dynamically changes along with ship states and meteorological conditions, so that the real scene characteristics of sea traffic are more fitted; The self-attention mechanism of the Patch-transform time sequence code is utilized to capture the cross-Patch length Cheng Shikong association, the coupling association is used as an attention weight correction term to optimize the association weight of a multi-ship coupling area, and finally the global space-time association matrix is corrected through iterative optimization, so that the accuracy and the comprehensiveness of space-time association recognition are effectively improved; Screening candidate associated edges according to a global space-time associated strength threshold, and filtering high-risk candidate edges through a comprehensive risk model (fusion of forward risk, starboard risk, draft risk and clearance risk); the candidate edges after depth-first search traversal screening are adopted, the comprehensive scores (the association strength is combined with the risk correction coefficient) of the paths are calculated, the paths with the scores larger than 0.85 are screened out to serve as key association paths, and the recognition result is ensured to have high association degree and navigation safety; The application suitability of the actual maritime scene is enhanced, namely the attribute (longitude, latitude, lower clearanc