A Method and Apparatus for Continuous Prediction of Ship Movement Probability Based on Behavioral Feature Maps

Abstract

This invention discloses a method and apparatus for continuous prediction of ship movement probabilities based on a behavioral feature graph. The method includes: acquiring the ship's sensitive regions, the importance coefficients of the sensitive regions, the interaction relationships between the ship and the sensitive regions, the interaction relationships between the sensitive regions, and the accompanying relationship features between ships; connecting the feature relationships to form a ship behavior knowledge graph containing multiple relationships between ships and land, with entities as nodes in the network structure, each node containing various ships and sensitive regions, and edges between nodes representing semantic relationships between entities; constructing basic data for ship movement prediction, predicting whether the relationship between the ship and the sensitive regions is close or sparse. This invention achieves the prediction of ship movement by constructing a ship movement feature knowledge graph, transforming direction prediction into a 9-direction transition probability, thereby predicting the probability change of ship presence.

Description

Technical Field

[0001] This invention belongs to the field of ship movement prediction technology, and particularly relates to a method and apparatus for continuous prediction of ship movement probability based on behavioral feature maps. Background Technology

[0002] With the widespread application of information sensing technologies such as shore-based / ship-based AIS, satellite remote sensing, space-based IoT, electromagnetic spectrum, and online public opinion monitoring, the observation data of ships at sea is becoming increasingly abundant, providing massive data support for ship movement analysis and prediction. By acquiring ship information from various sensing methods, analyzing ship behavior characteristics to obtain ship features, establishing a ship behavior knowledge graph, and constructing a continuous prediction model of ship movement based on this, it can be applied to multi-means continuous guidance and predicted position verification of key targets at sea, maintaining maritime emergency response and safety.

[0003] Currently, most ship movement prediction models predict discrete points in space and time, such as the non-patent literature "Ship Track Prediction Based on Sliding Window LSTM Network" (Wang Yukuan et al., Journal of Shanghai Maritime University, 043-1 (2022)). Based on the accumulation of observation data in one stage T0, the model predicts the trajectory points of the target in the next stage T1. This prediction is stage-based and discrete (the space is discrete, and the prediction result is a single point or multiple points; the time is also discrete, and the prediction result has time intervals), which leads to poor prediction results or inability to be applied in scenarios with high timeliness requirements (such as using data fusion of AIS, public opinion, electromagnetic spectrum and other means to guide space-based remote sensing means to continuously track and verify the position of the target).

[0004] Furthermore, existing methods for analysis and prediction mostly rely on data from a single source (primarily AIS), such as the non-patent literature "Research and Application of Ship Trajectory Analysis Based on AIS" (Yang Bochen. DISS. University of Electronic Science and Technology of China). This lack of utilization of multi-source data is problematic. The use of multi-source data requires high consistency across dimensions such as spatial scope, temporal scope, timeliness, resolution, and the observed object's form. For instance, before data fusion, the mainstream method extracts the target object from various data sources, clarifies its location and observation time, and then performs target matching and verification based on temporal and spatial overlap thresholds. This makes it difficult to fuse and match data with poor timeliness and low observation resolution with data with high timeliness and high observation resolution.

[0005] Furthermore, most current ship movement prediction methods directly model ship trajectory data, such as the non-patent literature "Application of a Parallel LSTM-FCN Model in Ship Track Prediction" (Hu Dan et al. Control and Decision. 37.8(2022):7). ​​These methods lack the influence of various factors on the ship's trajectory, such as waterways, key nodes, and destinations, which play an important role in the ship's navigation. At the same time, these methods often require normalization and standardization, which further weakens the temporal and spatial characteristics of the data. As a result, the model often constructs the "local" characteristics of the ship's navigation, lacking the modeling of "global" factors. Therefore, the prediction results will quickly fail. Summary of the Invention

[0006] In view of this, this invention designs a grid-based continuous prediction model for ship presence probability, transforming discrete prediction points into a continuous ship presence probability distribution, supporting continuous prediction of the target. As continuous validation observation data is "infused" into the model, the probability grid ensures continuous changes in the prediction results (location, time). It also designs a multi-source data-based ship behavior feature extraction method, introducing behavioral features and calculating the quantitative contribution of historical data to the current prediction by projecting ship information onto a map grid. This effectively solves the problems of difficulty in fusion due to inconsistent multi-dimensional observation scales and the high latency of real-time data acquisition leading to untimely application and data waste. Finally, it designs a ship behavior knowledge graph-based movement prediction method, which constructs a ship movement feature knowledge graph by mining sensitive areas in ship routes and the interaction relationships between ships and areas. Based on the knowledge graph, it achieves prediction of ship movement, enhancing the role of various elements in ship trajectory prediction during navigation.

[0007] The first aspect of this invention discloses a method for continuous prediction of ship movement probability based on behavioral feature maps, comprising the following steps:

[0008] Ship trajectory information is extracted from the data and projected onto a grid map. Spatial analysis is then performed on the map grid data to obtain the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction relationship between the ship and the sensitive areas, the interaction relationship between the sensitive areas, and the accompanying relationship characteristics between the ships.

[0009] By connecting the characteristics of sensitive areas of ships, the importance coefficient of sensitive areas, the interaction between ships and sensitive areas, the interaction between sensitive areas, and the accompanying relationship between ships, a ship behavior knowledge graph containing multiple relationships between ships and land is formed. Entities serve as nodes in the network structure, and nodes contain various ships and sensitive areas. The edges between nodes represent the semantic relationships between entities, including the interaction between sensitive areas, the interaction between ships and sensitive areas, and the accompanying relationship between ships.

[0010] The basic data for constructing ship movement prediction includes a ship dataset U and a sensitive area dataset V. Ship data is represented as u∈U, sensitive areas are represented as v∈V, and the interaction set of ship sensitive areas is represented as Y∈R. m×n Its physical meaning is an m×n matrix, representing the interaction relationships or correlation coefficients between m ships and n sensitive areas; predicting the interaction probability between ship u and sensitive area v. when and When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, when the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v. and A value close to 0 indicates a sparse relationship between the ship u and the sensitive area v.

[0011] Furthermore, projecting the ship trajectory information onto the grid map includes:

[0012] Ship trajectory information is extracted from various data such as space-based remote sensing, online public opinion, and AIS using CNN or Transformer algorithms. The ship information is projected onto a map grid. The ship trajectory information is discretized into latitude and longitude trajectory points and then the trajectory points are counted into the map grid corresponding to their latitude and longitude. Finally, the number of trajectory points in each grid is counted to obtain the commonly used waterways for ships.

[0013] The global land and sea map is divided into 1°×1° grids using latitude and longitude. All land areas are assigned a value of 0, and all sea areas and areas containing both land and sea surfaces are assigned a value of 1. The grid with actual geographic coordinates of 180°W and 90°N is defined as the top left vertex with grid coordinates (0, 0). Other grids are incremented sequentially to the right and downwards. Each grid is further numbered according to the rule that if the grid coordinates are (a, b), then the number is a×360+b. Based on the results obtained from the above rules, 1 represents connectivity and 0 represents disconnection, thus transforming the global map into a connected graph containing 360×180 nodes.

[0014] Furthermore, the calculation of the sensitive area includes: screening the high-frequency locations where ships appear, and using the k-nearest neighbor algorithm to cluster adjacent high-frequency locations to obtain multiple sensitive areas.

[0015] Furthermore, the calculation of the importance coefficient of sensitive areas includes: the frequency of visits by ships to different sensitive areas is different, and the importance coefficient is obtained based on the frequency of their entry and exit.

[0016] Furthermore, the interaction relationship between sensitive areas is calculated as follows: calculate the traffic flow of ships between sensitive areas. If a ship sails out of sensitive area A and enters sensitive area B, increase the interaction weight from A to B. Repeat this operation for the traffic flow data of all ships to obtain the interaction relationship between sensitive areas.

[0017] Furthermore, the interaction between ships and sensitive areas: based on the importance coefficient of sensitive areas and the interaction between sensitive areas, the visitation status of a specific ship in each sensitive area is calculated, and the interaction between ships and sensitive areas is obtained, supplementing the local characteristics of individual ships.

[0018] Furthermore, the companion relationship between ships is defined as follows: if n ships appear simultaneously within a region of radius R for a duration exceeding T, then these n ships have a companion relationship, and the strength of the companion relationship is determined by calculating the cumulative companion duration between the ships.

[0019] Furthermore, a triple structure is used to represent the knowledge graph of ship behavior, i.e., G = (E, R, S), where E represents the set of entities in the knowledge graph, including ships and sensitive areas; R is the set of relations in the knowledge graph, including frequent entries and exits, long-term companions, and close interactions; S represents the set of triples in the knowledge graph, and the form of S is represented by (entity, relation, entity), i.e., (h, r, t), where h is the head entity, r is the relation, and t is the tail entity.

[0020] The prediction function for ship movement is expressed as:

[0021]

[0022] in, Let θ represent the interaction probability between ship u and sensitive region v, θ represent the model parameters of the function, Y be the interaction matrix of the ship's sensitive region, and G be the ship behavior knowledge graph.

[0023] Furthermore, in the connected graph of the constructed grid, discrete information of various intelligences and predictions is projected onto a feature space composed of current position, destination, speed, and heading, based on the ship's size, speed, and draft characteristics. Based on these features and the prediction results based on the knowledge graph, the probability of the ship moving from the grid at time T0 to its adjacent grids above, below, left, right, upper left, upper right, lower left, and lower right, as well as the probability of it staying in the current grid, is obtained, i.e., the direction probability matrix. By multiplying the probability of the ship's current position by the direction probability matrix, the probability of the ship appearing in the grid at time T1 is obtained. This probability is then continuously diffused in nine directions to obtain the probability of the ship appearing from time T2 to time TN. In other words, the ship route with the highest probability is obtained in the connected graph, as well as the probability of the ship appearing in the grid at each time.

[0024] The second aspect of this invention discloses a continuous prediction device for ship movement probability based on behavioral feature maps, comprising:

[0025] Data extraction module: Extracts ship trajectory information from the data, projects the ship trajectory information onto a grid map, and then performs spatial analysis on the map grid data to obtain the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction relationship between the ship and the sensitive areas, the interaction relationship between the sensitive areas, and the accompanying relationship characteristics between the ships.

[0026] The knowledge graph construction module connects the features of sensitive areas of ships, the importance coefficient of sensitive areas, the interaction between ships and sensitive areas, the interaction between sensitive areas, and the accompanying relationship between ships to form a ship behavior knowledge graph that includes multiple relationships between ships and land. Entities serve as nodes in the network structure, and nodes contain various ships and sensitive areas. The edges between nodes represent the semantic relationships between entities, including the interaction between sensitive areas, the interaction between ships and sensitive areas, and the accompanying relationship between ships.

[0027] The ship movement prediction module: The basic data for constructing ship movement prediction includes a ship dataset U and a sensitive area dataset V. Ship data is represented as u∈U, sensitive areas are represented as v∈V, and the interaction set of ship sensitive areas is represented as Y∈R. m×n Its physical meaning is an m×n matrix, representing the interaction or correlation coefficients between m ships and n sensitive areas; when y uv ∈Y and y uv When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, y approaches 1. uv ∈Y and y uv A value close to 0 indicates a sparse relationship between the ship u and the sensitive area v.

[0028] The beneficial effects of this invention are as follows:

[0029] 1) By analyzing multi-source ship perception data, we can mine hidden sensitive areas and the interaction between ships and areas, and realize the construction of a knowledge graph based on ship movement characteristics.

[0030] 2) Construct a ship movement prediction model based on a ship movement feature knowledge graph to predict ship movements.

[0031] 3) Project the ship's navigation trajectory onto the grid diagram and convert the direction prediction into a 9-direction transition probability to predict the probability change of the ship's presence. Attached Figure Description

[0032] Figure 1 Schematic diagram of ship trajectory characteristics;

[0033] Figure 2 Flowchart of the ship movement probability prediction method based on behavioral feature maps of the present invention;

[0034] Figure 3Interaction feature map of sensitive regions;

[0035] Figure 4 The probability matrix at a certain moment in the grid distribution probability prediction results;

[0036] Figure 5 Navigation characteristics in the probability prediction results of grid distribution. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0038] The present invention discloses a method for continuous prediction of ship movement probability based on behavioral feature maps, comprising the following steps:

[0039] (I) Extraction of Ship Behavior Features Based on Multi-Source Data

[0040] Ship navigation trajectories contain a wealth of maritime route knowledge, including navigation patterns, commonly used routes, and behavioral characteristics of distinctive ships. This invention projects various data, such as AIS, space-based remote sensing, and online public opinion, onto a grid map, and then performs spatial analysis on the data to obtain features such as sensitive areas of ships, interaction relationships between ships and sensitive areas, interaction relationships between sensitive areas, and interaction relationships between ships, as detailed below:

[0041] 1. Multi-source data grid projection: Ship trajectory information is extracted from various data sources such as space-based remote sensing, online public opinion, and AIS using algorithms such as CNN and Transformer. This ship information is projected onto a map grid. By discretizing the ship trajectory information into latitude and longitude trajectory points and then counting these points into the corresponding map grids, the commonly used shipping routes can be obtained. Figure 1 .

[0042] 2. Sensitive Areas: Ships tend to spend more time at important locations during their missions (starting points, ending points, mission points, important turning points in the trajectory, etc.), which is reflected in the data as an increased frequency of occurrence. The frequency of these sensitive areas can be hundreds or thousands of times higher than that of ordinary areas. Based on this, high-frequency locations where ships appear are screened, and the k-nearest neighbor algorithm is used to cluster adjacent high-frequency locations, resulting in more than 40 sensitive areas.

[0043] 3. Importance coefficient of sensitive areas: The frequency of ship visits to different sensitive areas varies. Based on the frequency of ship entry and exit, the initial importance of the area can be determined. Figure 3 The horizontal axis represents the label of the sensitive area, and the vertical axis represents the frequency. The left side of the figure shows the statistics of ship entry frequency, and the right side shows the statistics of ship frequency.

[0044] 4. Interactions between sensitive areas: Sensitive areas are important nodes on the ship's trajectory. By calculating the traffic flow between sensitive areas, such as when a ship sails out of sensitive area A and enters sensitive area B, the interaction weight from A to B is increased. By repeating this operation on a large amount of data, the interaction relationships between sensitive areas can be obtained, which is used to support the construction of the knowledge graph in the following text and the calculation of the ship probability distribution.

[0045] 5. Ship-Sensitive Area Interaction: The importance coefficient of sensitive areas and the interaction between sensitive areas are global relationship parameters. They are obtained by performing probabilistic statistical analysis on a set of ship trajectories containing a large amount of data. These parameters include features that strengthen the collective characteristics of ship types and weaken the specificity of individual ships. Based on this, the visitation status of a specific ship in each sensitive area is calculated to obtain the interaction between the ship and the sensitive area, supplementing the local characteristics of individual ships.

[0046] 6. Accompanying Relationships Between Ships: In addition to their relationship with sensitive areas, ships also have accompanying relationships due to collaborative work and functional complementarity when performing tasks at sea. Therefore, if n ships appear simultaneously in an area with a radius of R for more than T hours, then these n ships are considered to have an accompanying relationship. The strength of the accompanying relationship is determined by calculating the cumulative accompanying duration between the ships.

[0047] (II) Construction of Ship Behavior Knowledge Graph and Trend Prediction

[0048] Predicting ship behavior requires analyzing historical ship behavior data to extract behavioral characteristics and combining these with spatiotemporal features to construct a spatiotemporal behavior graph of the ship. To this end, feature mining was performed on information from AIS behavior knowledge base, online public opinion knowledge base, and space-based intelligence knowledge base based on multi-source data ship behavior feature extraction. This yielded over 40 sensitive and active regions and their interrelationships. Based on this, ship and sensitive regions were further connected through relationships such as interactions between sensitive regions, interactions between ships and sensitive regions, and accompanying relationships between ships, forming a ship behavior knowledge graph encompassing various relationships between ships and land. This graph serves as the foundation for predicting maritime ship movements.

[0049] A ship behavior knowledge graph is a heterogeneous network topology with semantic information. Entities, serving as nodes in the network structure, include various ships, sensitive areas, etc. Edges between nodes represent semantic relationships between entities, including interactions between sensitive areas, interactions between ships and sensitive areas, and accompaniment relationships between ships. The knowledge graph is represented using a triple structure, G = (E, R, S), where E represents the set of entities in the knowledge graph (ships, sensitive areas, etc.); R is the set of relations in the knowledge graph (frequent entry / exit, long-term accompaniment, close interaction, etc.); and S represents the set of triples in the knowledge graph. S is represented in the form (entity, relation, entity), i.e., (h, r, t). Here, h is the head entity, r is the relation, and t is the tail entity. Entities, as node data in the knowledge graph, are the most basic elements of the knowledge graph, and edges between different entities represent relationships between them. Each entity can be represented by a globally unique ID.

[0050] The basic data for predicting ship movement includes a ship dataset U and a sensitive region dataset V. Ship data can be represented as u∈U, and sensitive regions as v∈V. The interaction set of ship-sensitive regions can be represented as Y∈R. m×n Physically, this is an m×n matrix representing the interaction relationships or correlation coefficients between m ships and n sensitive areas. It predicts the interaction probability between ship u and sensitive area v. when and When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, when the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v. and A value approaching 0 indicates a sparse relationship between the ship u and the sensitive region v. Therefore, the prediction function for ship movement is expressed as:

[0051]

[0052] Let θ represent the interaction probability between ship u and sensitive region v, θ represent the model parameters of the function, Y be the interaction matrix of the ship's sensitive region, and G be the ship behavior knowledge graph.

[0053] (III) Grid-based continuous prediction model for ship presence probability

[0054] Intelligence obtained from satellite remote sensing, AIS, and the Internet is often discrete. The prediction results of ship movement prediction models are destinations or directions of navigation. This information is scattered in the time dimension and lacks process information. In order to make discrete ship information continuous and make up for the lack of process information, it is necessary to construct a grid probabilistic prediction model based on ship characteristics.

[0055] First, the global land and sea map is divided into 1°×1° grids using a latitude and longitude grid. All land areas are assigned a value of 0, and all sea areas, as well as areas containing both land and sea surfaces, are assigned a value of 1. The grid with actual geographic coordinates of 180°W, 90°N is defined as the top-left vertex with grid coordinates (0, 0). Other grids are incremented sequentially to the right and downwards, as shown in the diagram. Each grid is further numbered according to the rule that if the grid coordinates are (a, b), then the number is a×360+b. Based on the above rules, 1 represents connectivity and 0 represents disconnection, thus transforming the global map into a connected graph containing 360×180 nodes.

[0056] After the connected graph of the grid is constructed, the discrete information of various intelligences and predictions is projected onto the feature space composed of the current position, destination, speed, and heading, based on the basic characteristics of the ship such as size, speed, and draft. Based on these features and the prediction results based on the knowledge graph, the probability of the ship moving from the current (T0) grid to its upper, lower, left, right, upper left, upper right, lower left, and lower right adjacent grids, as well as the probability of it staying in the current grid, can be obtained, i.e., the direction probability matrix (9-direction transfer probability). By multiplying the probability of the ship's current position (the probability at T0 is generally 1) by the direction probability matrix, the probability of the ship appearing in the grid at the next time (T1) can be obtained. This process is then repeated in nine directions to obtain the probability of the ship appearing at T2 up to TN. Thus, the ship route with the highest probability can be found in the connected graph, as well as the probability of the ship appearing in the grid at each time.

[0057] Based on the above process, the distribution matrix of the ship's appearance in the Earth's surface space at each moment can be calculated "procedurally" (e.g., Figure 4 ), and navigation characteristics in the absence of intelligence (such as Figure 5 ).

[0058] This invention projects the location and intelligence information of the target ship onto a grid, transforms the movement prediction results based on the knowledge graph into a direction probability matrix, and realizes the continuous prediction of the probability of ship appearance on the grid and the expression of spatial probability distribution through the direction probability matrix.

[0059] The present invention also discloses a continuous prediction device for ship movement probability based on behavioral feature maps, comprising:

[0060] Data extraction module: Extracts ship trajectory information from the data, projects the ship trajectory information onto a grid map, and then performs spatial analysis on the map grid data to obtain the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction relationship between the ship and the sensitive areas, the interaction relationship between the sensitive areas, and the accompanying relationship characteristics between the ships.

[0061] The knowledge graph construction module connects the features of sensitive areas of ships, the importance coefficient of sensitive areas, the interaction between ships and sensitive areas, the interaction between sensitive areas, and the accompanying relationship between ships to form a ship behavior knowledge graph that includes multiple relationships between ships and land. Entities serve as nodes in the network structure, and nodes contain various ships and sensitive areas. The edges between nodes represent the semantic relationships between entities, including the interaction between sensitive areas, the interaction between ships and sensitive areas, and the accompanying relationship between ships.

[0062] The ship movement prediction module: The basic data for constructing ship movement prediction includes a ship dataset U and a sensitive area dataset V. Ship data is represented as u∈U, sensitive areas are represented as v∈V, and the interaction set of ship sensitive areas is represented as Y∈R. m×n Its physical meaning is an m×n matrix, representing the interaction relationships or correlation coefficients between m ships and n sensitive areas; predicting the interaction probability between ship u and sensitive area v. when and When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, when the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v. and A value close to 0 indicates a sparse relationship between the ship u and the sensitive area v.

[0063] The beneficial effects of this invention are as follows:

[0064] 1) By analyzing multi-source ship perception data, we can mine hidden sensitive areas and the interaction between ships and areas, and realize the construction of a knowledge graph based on ship movement characteristics.

[0065] 2) Construct a ship movement prediction model based on a ship movement feature knowledge graph to predict ship movements.

[0066] 3) Project the ship's navigation trajectory onto the grid diagram and convert the direction prediction into a 9-direction transition probability to predict the probability change of the ship's presence.

[0067] As used herein, the term "preferred" is meant as an example, illustration, or illustration. Any aspect or design described herein as "preferred" need not be construed as being more advantageous than other aspects or designs. Rather, the use of the term "preferred" is intended to present the concept in a specific manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusionary "or." That is, unless otherwise specified or clear from the context, "X uses A or B" naturally includes either of the permutations. That is, if X uses A; X uses B; or X uses both A and B, then "X uses A or B" is satisfied in any of the foregoing examples.

[0068] Furthermore, although this disclosure has been shown and described with respect to one or more implementations, equivalent variations and modifications will occur to those skilled in the art based on a reading and understanding of this specification and the accompanying drawings. This disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the aforementioned components (e.g., elements, etc.), the terminology used to describe such components is intended to correspond to any component (unless otherwise indicated) that performs the specified function of said component (e.g., is functionally equivalent to it), even if structurally not equivalent to the disclosed structure performing the functions in the exemplary implementations of this disclosure shown herein. Moreover, although specific features of this disclosure have been disclosed with respect to only one of several implementations, such features may be combined with one or more features of other implementations that may be desirable and advantageous for a given or particular application. Furthermore, with regard to the use of the terms “comprising,” “having,” “containing,” or variations thereof in the Detailed Description or claims, such terms are intended to be included in a manner similar to the term “including.”

[0069] The functional units in this invention embodiment can be integrated into a processing module, or each unit can exist physically separately, or multiple units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. The aforementioned devices or systems can execute the storage methods in the corresponding method embodiments.

[0070] In summary, the above embodiments are one implementation of the present invention, but the implementation of the present invention is not limited to the embodiments described above. Any changes, modifications, substitutions, combinations, or simplifications made that deviate from the spirit and principle of the present invention should be considered equivalent substitutions and are included within the protection scope of the present invention.

Claims

1. A method for continuous prediction of ship movement probability based on behavioral feature maps, characterized in that, Includes the following steps: Ship trajectory information is extracted from the data and projected onto a grid map. Spatial analysis is then performed on the map grid data to obtain the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction relationship between the ship and the sensitive areas, the interaction relationship between the sensitive areas, and the accompanying relationship characteristics between the ships. The calculation of the sensitive area includes: screening the high-frequency locations where ships appear, and using the k-nearest neighbor algorithm to cluster adjacent high-frequency locations to obtain multiple sensitive areas; By connecting the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction between the ship and the sensitive areas, the interaction between the sensitive areas, and the accompaniment relationship between the ships, a ship behavior knowledge graph containing multiple relationships between the ship and the land is formed. Entities serve as nodes in the network structure, and each node contains various ships and sensitive areas. The edges between nodes represent the semantic relationships between entities, including the interaction between sensitive areas, the interaction between the ship and the sensitive areas, and the accompaniment relationship between the ships. The basic data for constructing ship movement prediction includes a ship dataset U and a sensitive area dataset V. The ship data is represented as follows: Sensitive areas are represented as The interaction set of ship sensitive areas is represented as Let be the correlation coefficients between m ships and n sensitive areas; predict the interaction probability between ship u and sensitive area v. ,when When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, when the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v. A value approaching 0 indicates a sparse relationship between the ship u and the sensitive area v; The prediction function for ship movement is expressed as: in, This represents the interaction probability between ship u and sensitive region v. The model parameters of the function are represented by Y, which is the interaction matrix of the ship's sensitive area, and G is the ship's behavior knowledge graph. In the connected graph of the constructed grid, discrete information of various intelligences and predictions is projected onto the feature space composed of current position, destination, speed, and heading, based on the ship's size, speed, and draft characteristics. Based on these features and the prediction results based on the knowledge graph, the probability of the ship moving from the grid at time T0 to its upper, lower, left, right, upper left, upper right, lower left, and lower right adjacent grids, as well as the probability of it staying in the current grid, is obtained, i.e., the direction probability matrix. By multiplying the probability of the ship's current position by the direction probability matrix, the probability of the ship appearing in the grid at time T1 is obtained. This probability is then continuously diffused in nine directions to obtain the probability of the ship appearing from time T2 to time TN. In other words, the ship route with the highest probability is obtained in the connected graph, as well as the probability of the ship appearing in the grid at each time.

2. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, The step of projecting the ship trajectory information onto a grid map includes: Ship trajectory information is extracted from various data such as space-based remote sensing, online public opinion, and AIS using CNN and Transformer algorithms. The ship information is projected onto a map grid. The ship trajectory information is discretized into latitude and longitude trajectory points and then the trajectory points are counted into the map grid corresponding to their latitude and longitude. Finally, the number of trajectory points in each grid is counted to obtain the commonly used waterways for ships. The global land and sea map is divided into 1°×1° grids using latitude and longitude. All land areas are assigned a value of 0, and all sea areas and areas containing both land and sea surfaces are assigned a value of 1. The grid with actual geographic coordinates of 180°W and 90°N is defined as the top left vertex with grid coordinates (0, 0). Other grids are incremented sequentially to the right and downwards. Each grid is further numbered according to the rule that if the grid coordinates are (a, b), then the number is a×360+b. Based on the results obtained from the above rules, 1 represents connectivity and 0 represents disconnection, thus transforming the global map into a connected graph containing 360×180 nodes.

3. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, The calculation of the importance coefficient of sensitive areas includes: the frequency of visits by ships to different sensitive areas is different, and the importance coefficient is obtained based on the frequency of their entry and exit.

4. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, The interaction relationship between sensitive areas is calculated as follows: Calculate the traffic flow of ships between sensitive areas. If a ship sails out of sensitive area A and enters sensitive area B, increase the interaction weight from A to B. Repeat this operation for the traffic flow data of all ships to obtain the interaction relationship between sensitive areas.

5. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, Ship-Sensitive Area Interaction: Based on the importance coefficient of sensitive areas and the interaction between sensitive areas, the visitation status of a specific ship in each sensitive area is calculated to obtain the interaction relationship between the ship and the sensitive area, thus supplementing the local characteristics of a single ship.

6. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, The companion relationship between ships is defined as follows: if n ships appear simultaneously within a region of radius R for a period of time exceeding T, then these n ships have a companion relationship. The strength of the companion relationship is determined by calculating the cumulative companion duration between the ships.

7. The method for continuous prediction of ship movement probability based on behavioral feature maps according to claim 1, characterized in that, The knowledge graph of ship behavior is represented using a triple structure, i.e., G = (E, R, S), where E represents the set of entities in the knowledge graph, including ships and sensitive areas; R is the set of relations in the knowledge graph, including frequent entries and exits, long-term companions, and close interactions; and S represents the set of triples in the knowledge graph. The form of S is (entity, relation, entity), i.e., (h, r, t), where h is the head entity, r is the relation, and t is the tail entity.

8. A continuous prediction device for ship movement probability based on behavioral feature maps, using the continuous prediction method for ship movement probability based on behavioral feature maps according to any one of claims 1-7, characterized in that, include: Data extraction module: Extracts ship trajectory information from the data, projects the ship trajectory information onto a grid map, and then performs spatial analysis on the map grid data to obtain the ship's sensitive areas, the importance coefficient of the sensitive areas, the interaction relationship between the ship and the sensitive areas, the interaction relationship between the sensitive areas, and the accompanying relationship characteristics between the ships. The calculation of the sensitive area includes: screening the high-frequency locations where ships appear, and using the k-nearest neighbor algorithm to cluster adjacent high-frequency locations to obtain multiple sensitive areas; The knowledge graph construction module connects the features of sensitive areas of ships, the importance coefficient of sensitive areas, the interaction between ships and sensitive areas, the interaction between sensitive areas, and the accompanying relationship between ships to form a ship behavior knowledge graph that includes multiple relationships between ships and land. Entities serve as nodes in the network structure, and nodes contain various ships and sensitive areas. The edges between nodes represent the semantic relationships between entities, including the interaction between sensitive areas, the interaction between ships and sensitive areas, and the accompanying relationship between ships. Ship movement prediction module: The basic data for constructing ship movement prediction includes a ship dataset U and a sensitive area dataset V. The ship data is represented as follows: Sensitive areas are represented as The interaction set of ship sensitive areas is represented as Its physical meaning is an m×n matrix, representing the interaction relationships or correlation coefficients between m ships and n sensitive areas; predicting the interaction probability between ship u and sensitive area v. ,when When the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v; conversely, when the value approaches 1, it indicates a close relationship between the ship u and the sensitive area v. A value approaching 0 indicates a sparse relationship between the ship u and the sensitive area v; The prediction function for ship movement is expressed as: in, This represents the interaction probability between ship u and sensitive region v. The model parameters of the function are represented by Y, which is the interaction matrix of the ship's sensitive area, and G is the ship's behavior knowledge graph. In the connected graph of the constructed grid, discrete information of various intelligences and predictions is projected onto the feature space composed of current position, destination, speed, and heading, based on the ship's size, speed, and draft characteristics. Based on these features and the prediction results based on the knowledge graph, the probability of the ship moving from the grid at time T0 to its upper, lower, left, right, upper left, upper right, lower left, and lower right adjacent grids, as well as the probability of it staying in the current grid, is obtained, i.e., the direction probability matrix. By multiplying the probability of the ship's current position by the direction probability matrix, the probability of the ship appearing in the grid at time T1 is obtained. This probability is then continuously diffused in nine directions to obtain the probability of the ship appearing from time T2 to time TN. In other words, the ship route with the highest probability is obtained in the connected graph, as well as the probability of the ship appearing in the grid at each time.

Images