Method for generating marine navigation network based on vector-raster model fusion

Abstract

The application provides a marine navigation network generation method based on vector-grid model fusion, and relates to the technical field of waterway autonomous route design.The steps of the method include obtaining original AIS track data of different ship types during marine navigation;for each ship type, identifying a berthing track section, extracting a voyage track, and performing abnormal track identification and high-quality route extraction to obtain a representative route; a quadtree spatial grid model is constructed, the representative route is segmented and topologically reconstructed at multiple precisions based on in-out grid anchor points to obtain a target route segment, a vector navigation network is constructed through spatial and heading clustering, a grid navigation network is constructed, a route point matching strategy is introduced, a topological association between the vector and grid routes is established, and a vector-grid navigation network is generated; route arc segments of different ship types are extracted, a matching model is constructed for arc segment matching, and route fusion and incremental reconstruction are performed on the vector-grid navigation network to generate a multi-ship-type fusion marine navigation network.

Description

Technical Field

[0001] This invention relates to the field of autonomous waterway route design technology, and in particular to a method for generating maritime navigation networks based on vector-grid model fusion. Background Technology

[0002] With the deepening of trade globalization, waterway transportation, with its capacity to carry large quantities of goods and its cost advantages over long distances, occupies an important position in the international freight transport system. Route planning, as a crucial link in maritime transport, plays a vital supporting role in enabling ships to cope with complex marine environments, ensure navigational safety, and improve transport efficiency. Currently, ship route planning is gradually evolving from traditional experience-based decision-making towards intelligent and autonomous systems, placing higher demands on the accuracy, reliability, and spatial coverage of maritime navigation networks.

[0003] Existing methods for constructing maritime navigation networks typically rely on historical AIS trajectory data of ships. By mining typical navigation patterns of ships, a navigation network organized in a node-edge graph structure is built, thus providing historical experience references for route planning. However, due to the discreteness of the spatial distribution of ship AIS trajectories and the differences in navigation patterns among different ship types, navigation networks built solely based on the historical trajectories of a single ship type not only suffer from insufficient spatial accessibility in waters not covered by ships, but also experience insufficient connectivity between network nodes in covered waters due to the limitations of single-ship-type data. This makes it difficult to simultaneously meet the comprehensive requirements of route planning across multiple ship types and scales in terms of network accuracy, spatial accessibility, and topological connectivity. Summary of the Invention

[0004] The purpose of this invention is to provide a method for generating maritime navigation networks based on vector-grid model fusion, in order to solve the problem mentioned in the background art that existing maritime navigation network construction methods are difficult to simultaneously meet the comprehensive requirements of route planning for network accuracy, spatial accessibility and topological connectivity in waters with multiple ship types and scales.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for generating a maritime navigation network based on vector-grid model fusion, comprising the following steps: acquiring raw AIS trajectory data of different ship types during maritime navigation and preprocessing it; for each ship type, identifying berthing trajectory segments from the preprocessed AIS trajectory data based on port dwell time, extracting continuous trajectories between adjacent different ports of call as voyage trajectories, and performing abnormal trajectory identification and high-quality route extraction on the voyage trajectories to obtain representative routes; constructing a quadtree spatial grid model using known offshore distances and port distribution, and using the entry and exit grid anchor points in the quadtree spatial grid model to generate the representative routes. The target route segment is obtained by performing multi-precision segmentation and topology reconstruction on the target route segment. A vector navigation network is constructed by spatial and directional clustering of the target route segment. A grid navigation network is constructed based on land boundary constraints and triangulation methods. A waypoint matching strategy based on node connectivity and spatial similarity is introduced to establish the topological association between vector and grid routes, generating a vector-grid navigation network. Route arc segments are extracted from the vector-grid navigation networks corresponding to different ship types. A matching model based on direction, length, and geospatial similarity is constructed for arc segment matching. The vector-grid navigation network is then fused and incrementally reconstructed to generate a multi-ship type fused maritime navigation network.

[0006] Optionally, the step of identifying berthing trajectory segments specifically includes: detecting trajectory points located within the same port area, identifying continuous trajectory points with a dwell time span greater than a preset time threshold as berthing trajectory segments, and marking the ports where the trajectory points are located as berthing ports; traversing the berthing trajectory segments in chronological order, and if two adjacent berthing trajectory segments berth at different ports, extracting the trajectory points between the end point of the previous berthing trajectory segment and the start point of the next berthing trajectory segment as a voyage trajectory segment.

[0007] Optionally, the steps of identifying abnormal trajectories and extracting high-quality routes for the voyage trajectory specifically include: performing cubic spline interpolation and smoothing on the voyage trajectory; for the smoothed voyage trajectory, minimizing the sum of the penalty terms for the squared error of the heading angle and the number of feature points within the segment, and introducing an evaluation index based on the number of feature points and Hausdorff distance to adaptively optimize the penalty parameters in the penalty terms to extract turning feature points; simultaneously, for the smoothed voyage trajectory, constructing a sliding window, detecting the distribution of inflection points in the longitude and latitude directions of the trajectory within the window, calculating the redundancy of the turning-around type wandering trajectory and the redundancy of the looping type wandering trajectory respectively, and removing the corresponding abnormal trajectory segments when the redundancy exceeds a preset redundancy threshold; for the voyage trajectory after removing abnormal trajectory segments, voyage trajectories with the same starting port and destination port are grouped together, using Hausdorff distance as a similarity index, hierarchical clustering is adopted and a silhouette coefficient is introduced to determine the globally optimal distance threshold, and the trajectory with the shortest voyage in each cluster is extracted as the representative route.

[0008] Optionally, the step of constructing a quadtree spatial grid model using known offshore distances and port distribution specifically includes: mapping the global sea area onto the six projection planes of a cube using a cube projection method, constructing regular grids on each projection plane to form a basic grid model; traversing all grids, calculating the offshore scale ratio of each grid based on the closest distance of the grid region to the coastline and the current grid side length, and counting the number of ports in each grid; within the maximum number of iterations, if the number of ports in the current grid is 0 and the offshore scale ratio is greater than the maximum offshore scale ratio, then the segmentation process of the current grid ends; otherwise, quadtree segmentation is performed on the current grid, and the segmented sub-grids are processed recursively.

[0009] Optionally, the step of performing multi-precision segmentation and topology reconstruction of the representative route based on the entry and exit grid anchor points in the quadtree spatial grid model specifically includes: determining the grid sequence traversed by the route according to the positional relationship between the route nodes and the grid, and extracting the intersection points of the route and each grid boundary as entry and exit grid anchor points; determining the perpendicular bisectors of adjacent entry and exit grid anchor points, and using the perpendicular bisectors as the segmentation baseline to divide the representative route into several discrete segments, wherein the two ends of the segment are segmentation anchor points, and the turning feature points of the route are retained in the middle; and constructing a directed adjacency matrix of the segments according to the adjacency relationship of the segments in the same route.

[0010] Optionally, the step of constructing a vector navigation network by spatial and heading clustering of the target route segments specifically includes: using a two-layer sorting point identification clustering structure algorithm, sorting the spatial density of the route segments in the spatial layer to generate a reachability map based on Hausdorff distance, and identifying routes exceeding a preset steepness threshold as cluster boundaries by calculating the steepness of the similarity changes between adjacent routes in the reachability map, and segmenting the reachability map to obtain spatial clusters of routes; in the heading layer, for each spatial cluster, calculating the average heading angle of the route, using the difference between the cumulative sum of the turns of two routes as a heading similarity measure, constructing a reachability map in the heading dimension, and extracting the final cluster from the reachability map using a steepness-based clustering extraction method; calculating the spatial centrality score of each route in the final cluster, selecting the route with the highest score as the representative route, extracting the feature points of the representative route as waypoints, and mapping the waypoints to routes according to the directed connection relationships between waypoints.

[0011] Optionally, the step of constructing a grid navigation network based on land boundary constraints and triangulation specifically includes: extracting grid edges and diagonals as grid routes based on the quadtree spatial grid model to construct an initial grid navigation network; introducing a quadtree hierarchical partitioning strategy to refine the initial grid navigation network at multiple scales, using the Alpha shape algorithm to generate the concave hull boundary of the refined initial grid navigation network, and calculating the global water area coverage rate covered by the concave hull boundary; if the coverage rate is lower than a preset coverage rate threshold, then performing the next level of quadtree partitioning on the uncovered water areas until the coverage rate is not lower than the preset coverage rate threshold; for waypoints whose connectivity decreases due to land boundary constraints, they are identified as degraded waypoints, and a Delaunay triangulation is constructed based on the set of degraded waypoints, retaining the triangular edges that do not intersect with the original grid routes and land boundaries as new routes.

[0012] Optionally, the step of introducing a waypoint matching strategy based on node connectivity and spatial similarity to establish the topological association between vector and grid waypoints specifically includes: performing matching calculations on vector waypoints within a neighborhood of a grid waypoint radius of τ, calculating the comprehensive matching degree between each vector waypoint and the current grid waypoint, selecting the vector waypoint with the highest comprehensive matching degree, updating the coordinates of the grid waypoint with weighted average, and synchronously updating the geometric spatial coordinates of all associated waypoints.

[0013] Optionally, the step of constructing a matching model based on direction, length, and geospatial similarity for arc segment matching specifically includes: taking the number of routes heading towards and out of a waypoint as the in-degree and out-degree of the current waypoint, respectively; extracting linear route sequences between waypoints with an in-degree or out-degree greater than 1 as route arc segments; performing grid encoding on the route arc segments to obtain the grid sequence corresponding to each route arc segment; extracting route arc segments with common grid subsequences as associated route arc segment pairs; calculating the direction similarity, length similarity, and geospatial similarity of the associated route arc segment pairs; fusing the direction and length geometric similarity features through weighted summation; and introducing geographical environment constraints as binary gate variables to calculate the comprehensive matching degree. If the comprehensive matching degree is not less than a preset matching degree threshold, the arc segment pair is considered a match.

[0014] Optionally, the steps of performing route fusion and incremental reconstruction on the vector-grid navigation network specifically include: traversing the route arc matching matrix; for fully matched arc pairs, performing weighted average fusion on the coordinates of waypoints with the same index in the arc pair; for partially matched arc pairs, extracting continuous sub-arc pairs with the same matching relationship, and performing weighted average fusion on the coordinates of waypoints with the same index in the matched sub-arc pairs; performing weighted average calculation on the coordinates of the starting and ending waypoints of mismatched sub-arcs respectively, so as to insert the starting and ending waypoints shared by the mismatched and matched sub-arcs, and retaining the original coordinates and topology of the remaining mismatched sub-arcs as new routes for merging and fusion; and retaining the original coordinates and topology of mismatched arc pairs as new routes for merging and fusion.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This application effectively overcomes the problem of insufficient node connectivity caused by single-ship-type data by extracting route arcs from the vector-grid navigation networks corresponding to different ship types and performing arc matching and fusion based on direction, length, and geospatial similarity. This significantly improves the number of reachable ports and route connectivity of the network in multi-ship-type scenarios. At the same time, by topologically associating and fusing the vector navigation network built based on historical trajectories with the grid navigation network built based on land boundary constraints and triangulation, it not only preserves the actual navigation experience of ships but also makes up for the spatial blind spots in waters not covered by the trajectory. This allows the generated maritime navigation network to maintain high accuracy in high-risk nearshore waters and adaptively reduce complexity in low-risk open ocean waters. Thus, it balances the integrity of the global space, the reliability of navigation experience, and the computational efficiency of route planning, providing a reliable spatial pattern expression for route planning in multi-ship-type and multi-scale waters. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0017] Figure 2 This is a schematic diagram of the ship's turning-around and circling hovering scenarios according to the present invention.

[0018] Figure 3 This is a schematic diagram of the ship route reconstruction process of the present invention.

[0019] Figure 4 This is a schematic diagram of the vector navigation network construction of the present invention.

[0020] Figure 5 This is a schematic diagram of the grid navigation network enhancement of the present invention.

[0021] Figure 6 This is a schematic diagram of the vector-grid navigation network fusion of the present invention.

[0022] Figure 7 This is a schematic diagram illustrating the fusion of fully matched, partially matched, and mismatched flight path arc pairs according to the present invention. Detailed Implementation

[0023] The present invention will now be clearly and completely described in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0026] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0027] It should be understood that the sequence number and size of each step in this embodiment do not imply the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.

[0028] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0029] Please refer to Figures 1-7 This invention discloses a method for generating a maritime navigation network based on vector-grid model fusion. The steps include: acquiring raw AIS trajectory data of different ship types during maritime navigation and preprocessing it; for each ship type, identifying berthing trajectory segments from the preprocessed AIS trajectory data based on port dwell time, extracting continuous trajectories between adjacent ports of call as voyage trajectories, and performing abnormal trajectory identification and high-quality route extraction on the voyage trajectories to obtain representative routes; constructing a quadtree spatial grid model using known offshore distances and port distribution, and performing multi-precision analysis on the representative routes based on the entry and exit grid anchor points in the quadtree spatial grid model. Degree segmentation and topology reconstruction are performed to obtain target route segments. Spatial and directional clustering is used to construct a vector navigation network for the target route segments. A grid navigation network is constructed based on land boundary constraints and triangulation methods. A waypoint matching strategy based on node connectivity and spatial similarity is introduced to establish the topological association between vector and grid routes, generating a vector-grid navigation network. Route arc segments are extracted from the vector-grid navigation networks corresponding to different ship types. A matching model based on direction, length, and geospatial similarity is constructed to perform arc segment matching. Route fusion and incremental reconstruction are performed on the vector-grid navigation network to generate a multi-ship type fused maritime navigation network.

[0030] Specifically, raw AIS trajectory data of different ship types (such as container ships, tankers, bulk carriers, etc.) during their navigation at sea are obtained from shore-based base stations or satellite receiving stations of the Automatic Identification System (AIS).

[0031] The acquired raw AIS data is preprocessed, including removing obvious outliers (such as latitude and longitude exceeding the normal range, speed exceeding the ship's maximum design speed, etc.), filling missing data segments using linear interpolation or cubic spline interpolation methods, and uniformly formatting the timestamps in the AIS data.

[0032] For each ship type, berthing trajectory segments are identified based on the ship's dwell time in the port. By calculating whether the ship's trajectory points are spatially located within the port buffer zone and statistically analyzing the time span of consecutive trajectory points within the port zone, trajectory point sequences with a time span exceeding a preset threshold are marked as berthing trajectory segments. Continuous trajectories between two adjacent ports of call are extracted as voyage trajectories. For example, if the port corresponding to the previous berthing trajectory segment is A, and the port corresponding to the next berthing trajectory segment is B, then the trajectory from the berthing end point at port A to the berthing start point at port B constitutes a voyage trajectory from A to B. Subsequently, cubic spline interpolation and Savitzky-Golay filtering are used to interpolate and smooth each voyage trajectory segment.

[0033] For each voyage trajectory, abnormal trajectory identification is performed, including detecting and eliminating trajectory segments caused by signal loss, multipath effect or temporary vessel wandering, and extracting the highest quality trajectory from multiple voyage trajectories between the same port pairs as the representative route.

[0034] Offshore distances are calculated using known global coastline data, and a quadtree spatial grid model is constructed by combining it with port distribution data. The grid resolution is dynamically adjusted based on the offshore distance and port density. Based on the entry and exit grid anchor points (i.e., the intersections where the route enters and exits each grid boundary) in the quadtree spatial grid model, representative routes are segmented with multi-precision resolution. This results in higher spatial resolution for routes in nearshore high-port-density areas and lower resolution in offshore low-port-density areas. The segmented routes are then topologically reconstructed to form target route segments.

[0035] Vector networks and grid networks are constructed for the target route segments. In the vector network construction, a combination of spatial clustering and heading clustering is used to aggregate segments with similar spatial locations and consistent headings into routes, extract waypoints, and establish directed connections. In the grid network construction, based on land boundary constraints and the Delaunay triangulation method, grid edges and diagonals in the quadtree grid model are filtered and supplemented to form a connected and obstacle-avoiding grid navigation network.

[0036] A waypoint matching strategy based on node connectivity and spatial similarity is introduced to calculate the comprehensive similarity between vector waypoints and grid waypoints in terms of spatial location and topology. The successfully matched waypoints are then fused by coordinate weighting to establish a topological association between the vector network and the grid network, thereby generating a unified vector-grid navigation network.

[0037] Extract route arcs (i.e., continuous waypoint sequences) from the vector-grid navigation networks corresponding to different ship types, construct a matching model based on heading direction, arc length, and geospatial overlap, perform similarity matching on route arcs from different ship types, merge successfully matched arcs, and incrementally add unmatched arcs. By iteratively updating the network topology, a multi-ship type fusion maritime navigation network that can simultaneously adapt to the navigation needs of multiple ship types is generated.

[0038] This application effectively overcomes the problem of insufficient node connectivity caused by single-ship-type data by extracting route arcs from the vector-grid navigation networks corresponding to different ship types and performing arc matching and fusion based on direction, length, and geospatial similarity. This significantly improves the number of reachable ports and route connectivity of the network in multi-ship-type scenarios. At the same time, by topologically associating and fusing the vector navigation network built based on historical trajectories with the grid navigation network built based on land boundary constraints and triangulation, it not only preserves the actual navigation experience of ships but also makes up for the spatial blind spots in waters not covered by the trajectory. This allows the generated maritime navigation network to maintain high accuracy in high-risk nearshore waters and adaptively reduce complexity in low-risk open ocean waters. Thus, it balances the integrity of the global space, the reliability of navigation experience, and the computational efficiency of route planning, providing a reliable spatial pattern expression for route planning in multi-ship-type and multi-scale waters.

[0039] In some embodiments, the step of identifying berthing trajectory segments specifically includes: detecting trajectory points located within the same port area, identifying continuous trajectory points with a dwell time span greater than a preset time threshold as berthing trajectory segments, and marking the ports where the trajectory points are located as berthing ports; traversing the berthing trajectory segments in chronological order, and if two adjacent berthing trajectory segments berth at different ports, extracting the trajectory points between the end point of the previous berthing trajectory segment and the start point of the next berthing trajectory segment as a voyage trajectory segment.

[0040] Specifically, the location coordinates of global ports and the port buffer radius (usually set to 5 to 10 nautical miles) are obtained. For each trajectory point in the preprocessed AIS trajectory data, the distance from the point to the center point of each port is calculated. If the distance is less than the buffer radius of the port, the trajectory point is determined to be within the range of the port.

[0041] Trajectory points are traversed in chronological order, and consecutive trajectory points located within the same port area are grouped into candidate berthing segments. The timestamp difference between the first and last trajectory points in a candidate berthing segment is calculated as the dwell time span. If this span is greater than a preset time threshold (e.g., 1 hour for container ships and 2 hours for tankers, which can be dynamically adjusted according to ship type), the candidate berthing segment is identified as a formal berthing trajectory segment, and the ports of all trajectory points within this segment are marked as the corresponding ports of call.

[0042] When extracting the voyage trajectory, all identified berthing trajectory segments are arranged in chronological order. The berthing trajectory segments are then traversed sequentially, and adjacent berthing trajectory segments are checked one by one. If adjacent berthing trajectory segments... and If the ports of call are different, for example, the first segment calls at port P and the second segment calls at port Q, then the trajectory point with the latest time in the first segment is used. The earliest track point in the subsequent parking track segment Extract all trajectory points between ports. The continuous trajectory formed by these trajectory points constitutes a complete voyage trajectory from port P to port Q. If two adjacent berthing trajectory segments stop at the same port, it is considered that the ship has stopped at that port multiple times without changing its berthing status. In this case, the voyage trajectory is not extracted, and the process continues to traverse forward.

[0043] This application identifies berthing trajectory segments based on port stay time and extracts continuous trajectories between different ports of call as voyage trajectories in chronological order. It can automatically and accurately segment the original AIS trajectory into independent voyages with clearly defined start and end ports, effectively avoiding missegmentation caused by short-term vessel movement within ports or signal jitter. This ensures the integrity and semantic correctness of the voyage trajectory, provides clean and structured input data for subsequent abnormal trajectory identification and high-quality route extraction, and improves the automation and reliability of navigation network generation.

[0044] In some embodiments, the steps of identifying abnormal trajectories and extracting high-quality routes for the voyage trajectory specifically include: performing cubic spline interpolation and smoothing on the voyage trajectory; for the smoothed voyage trajectory, minimizing the sum of the penalty terms for the squared error of the heading angle and the number of feature points within the segments, and introducing an evaluation index based on the number of feature points and Hausdorff distance to adaptively optimize the penalty parameters in the penalty terms to extract turning feature points; simultaneously, for the smoothed voyage trajectory, constructing a sliding window, detecting the distribution of inflection points in the longitude and latitude directions of the trajectory within the window, calculating the redundancy of the turning-around type wandering trajectory and the redundancy of the looping type wandering trajectory respectively, and removing the corresponding abnormal trajectory segments when the redundancy exceeds a preset redundancy threshold; for the voyage trajectory after removing abnormal trajectory segments, voyage trajectories with the same starting port and destination port are grouped together, using Hausdorff distance as a similarity index, hierarchical clustering is adopted and a silhouette coefficient is introduced to determine the globally optimal distance threshold, and the trajectory with the shortest voyage in each cluster is extracted as the representative route.

[0045] Specifically, cubic spline interpolation is performed on the longitude, latitude, and time series of each voyage trajectory. The interpolated time intervals are standardized to a fixed value (e.g., 30 seconds). Then, a Savitzky-Golay filter is used for smoothing to eliminate high-frequency noise and short-term fluctuations. For the smoothed voyage trajectory, feature points that can characterize significant ship maneuvers are extracted to preserve key directional changes in the route.

[0046] The trajectory point sequence is divided into several sub-intervals. By minimizing the sum of the squared error of the heading angle and the penalty term for the number of changing points within each segment, a balance is achieved between model complexity and fitting accuracy. Penalty parameter. The accuracy of feature point extraction is determined when When the value is large, the model tends to reduce the number of feature points, thus obtaining a smoother compression result; conversely, when the value is small, the model tends to reduce the number of feature points. When the steering angle is small, it is more sensitive to small steering angles and is prone to introducing redundant steering feature points.

[0047] By introducing feature point evaluation indicators ,accomplish Adaptive tuning. Specifically, firstly, a large initial penalty parameter is set. According to a fixed step size The penalty parameter is gradually reduced, and the PELT algorithm is used to extract steering feature points under adjacent penalty parameters, and the corresponding feature point evaluation index is calculated. and .like Then continue with the step size The penalty parameter is decreased for each iteration until the feature point evaluation index no longer decreases. Then, the penalty parameter and its corresponding feature point set from the previous iteration are used as the final output. The formulas for calculating the objective function and the feature point evaluation index are:

[0048] ; ; in, Represents the set of feature points. express The heading angle of the feature point at that moment. For the first The mean of the heading angles of feature points within each segment. The number of feature points, Here, n is the penalty parameter, and n is the number of extracted turning feature points. This represents the Hausdorff distance between the original trajectory and the processed trajectory.

[0049] Due to factors such as navigation control and port entry queues, some vessel voyage tracks exhibit abnormal segments with a back-and-forth lingering spatial pattern. These abnormal track segments can interfere with the route extraction results, leading to distortion of the navigation network structure. This application analyzes the variation characteristics of vessel tracks along latitude and longitude lines, classifying lingering tracks into two types: turning-around and circular. The former is characterized by monotonic variation along one coordinate axis, while exhibiting frequent oscillations along its orthogonal axis, with the overall track still moving slowly along another coordinate axis; the latter is characterized by non-monotonic variation along both coordinate axes, with the overall track exhibiting circular or closed characteristics, and the activity range remaining relatively stable.

[0050] A sliding window is constructed for the smoothed flight trajectory, with the window size set to 5%–10% of the total number of points on the trajectory. Within each window, the variation components of the trajectory vector along the longitude and latitude axes are extracted, and the inflection point distribution of trajectory points in the longitude and latitude directions is detected: inflection points in the longitude direction are defined as points where the longitude value changes from increasing to decreasing or vice versa, and inflection points in the latitude direction are similar. If only one dimension of longitude or latitude has an inflection point, there may be a zigzag trajectory segment, and the redundancy of the zigzag trajectory is calculated. The redundancy of the turnaround trajectory characterizes the minimum number of heading turns included in the turnaround lingering trajectory within the sliding window. Let the redundancy threshold be... ,like If there are inflection points in both longitude and latitude, then there may be looping trajectory segments within the window. Calculate the redundancy of the looping trajectory. The redundancy of the circling trajectory characterizes the minimum number of circling maneuvers contained in the circling trajectory within the sliding window. Let the circling circling redundancy threshold be... ,if If so, then there is a circling trajectory segment within the window. The calculation formula is:

[0051] ; ; ; In the formula: Redundancy of the zigzag trajectory. This represents the total number of trajectory points within the sliding window. This represents the number of trajectory points along the longitude direction. This represents the number of trajectory points along the latitudinal direction. This refers to the redundancy of the loop trajectory.

[0052] The circling or turning-around trajectory segments corresponding to the window are identified as abnormal trajectories and removed. For the remaining voyage trajectories after removing abnormal trajectories, they are grouped according to the origin and destination ports, with all voyage trajectories within each group sharing the same origin and destination port pair. Using Hausdorff distance as the similarity index between trajectories, a hierarchical clustering algorithm is employed to cluster the voyage trajectories within each group. During the clustering process, the silhouette coefficient is used to evaluate the clustering effect under different distance thresholds, and the distance threshold that maximizes the average silhouette coefficient is selected as the globally optimal distance threshold. Within each cluster, the voyage distance for each trajectory is calculated, and the trajectory with the shortest voyage distance is selected as the representative route for that cluster. The representative route reflects the most economical or frequently used navigation path between that port pair.

[0053] This application effectively filters out low-quality trajectory fragments caused by signal noise, temporary vessel wandering, or abnormal route detours by interpolating and smoothing voyage trajectories, extracting adaptive turning feature points, and identifying and eliminating wandering trajectories based on a sliding window. By combining hierarchical clustering and adaptively determining clustering thresholds using silhouette coefficients, and selecting the trajectory with the shortest distance in each cluster as the representative route, the most economical and typical navigation paths can be selected from multiple voyages originating and terminating at the same port. This ensures the cleanliness of the representative routes while preserving the differences between different navigation modes, providing high-quality basic route data for subsequent network construction.

[0054] In some embodiments, the step of constructing a quadtree spatial grid model using known offshore distances and port distribution specifically includes: mapping the global sea area onto the six projection planes of a cube using a cube projection method, constructing regular grids on each projection plane to form a basic grid model; traversing all grids, calculating the offshore scale ratio of each grid based on the closest distance of the grid region to the coastline and the current grid side length, and counting the number of ports in each grid; within the maximum number of iterations, if the number of ports in the current grid is 0 and the offshore scale ratio is greater than the maximum offshore scale ratio, then the segmentation process of the current grid ends; otherwise, quadtree segmentation is performed on the current grid, and the segmented sub-grids are processed recursively.

[0055] Specifically, the cube projection method is used to map the global ocean area (longitude -180° to 180°, latitude -90° to 90°) onto six projection faces of a cube, with each projection face corresponding to a square region. On each projection face, regular rectangular grids are constructed according to preset initial grid side lengths (e.g., the arc length corresponding to 1°×1° at the equator), thus forming a basic grid model covering the entire globe.

[0056] Traverse all grid cells in the base grid model. For each grid cell, calculate the shortest distance from its geometric center point to the nearest coastline, denoted as the offshore distance. Divide this offshore distance by the actual side length of the current grid cell to obtain the offshore scale ratio, calculated using the following formula: ;in, This indicates the closest distance of the grid area to the coastline. This indicates the current grid side length.

[0057] Count the number of ports falling within each grid cell. Set a maximum number of iterations (e.g., 10). For each grid cell, check if the segmentation termination condition is met: if the number of ports in the current grid cell is 0 and the offshore scale ratio is greater than the preset maximum offshore scale ratio (e.g., 5.0, indicating that the grid cell is located in the open ocean and has no ports), then terminate further segmentation of the grid cell and mark it as a leaf node; otherwise, perform quadtree segmentation on the grid cell, that is, divide the grid cell into four sub-grids, and recursively repeat the above judgment and segmentation process for each sub-grid.

[0058] Through the above recursive segmentation, densely populated nearshore areas and smaller offshore sea areas are divided into finer grids, while open ocean areas without ports maintain a larger grid size, thus achieving an adaptive multi-resolution representation of global sea areas.

[0059] This application employs a cubespan projection method to map the global ocean area onto six projection planes, avoiding the severe deformation of traditional latitude and longitude grids in high-latitude regions and ensuring the relative uniformity of grid cell geometry globally. By introducing offshore scale ratio and port number as termination conditions for quadtree segmentation, the grid is automatically refined in densely populated nearshore areas and maintains coarse granularity in portless open ocean areas, achieving adaptive multi-resolution representation of the ocean space. This reduces storage and computational overhead in open ocean areas while ensuring the accuracy of navigation networks in complex nearshore waters, laying a solid spatial foundation for subsequent multi-precision route segmentation and grid network construction.

[0060] In some embodiments, the step of performing multi-precision segmentation and topology reconstruction of the representative route based on the entry and exit grid anchor points in the quadtree spatial grid model specifically includes: determining the grid sequence traversed by the route according to the positional relationship between the route nodes and the grid, and extracting the intersection points of the route and each grid boundary as entry and exit grid anchor points; determining the perpendicular bisectors of adjacent entry and exit grid anchor points, and using the perpendicular bisectors as the segmentation baseline to divide the representative route into several discrete segments, wherein the two ends of the segment are segmentation anchor points, and the turning feature points of the route are retained in the middle; and constructing a directed adjacency matrix of the segments according to the adjacency relationship of the segments in the same route.

[0061] Specifically, each representative flight path is represented as a series of consecutive latitude and longitude coordinates. Based on the quadtree grid containing each coordinate point (determined by mapping the coordinates to a cube projection and traversing the quadtree), the sequence of grids traversed by the flight path is determined, i.e., a list of grid identifiers passed sequentially in the flight order. The intersections between the flight path and the boundaries of each grid are extracted; these intersections are the grid entry and exit anchor points. Each anchor point records the position coordinates of the flight path entering or exiting an adjacent grid. Then, all adjacent grid entry and exit anchor point pairs are traversed. For each pair of adjacent anchor points, the perpendicular bisector of the line connecting the two anchor points is calculated, and this perpendicular bisector is used as the dividing baseline.

[0062] The original representative route is divided into several discrete segments using the intersection of the segmented baseline and the original route trajectory as the segmentation point. Each segment has segmentation anchor points at both ends (i.e., grid entry and exit anchor points), while the original turning feature points are retained in the middle of the segment. This segmentation method ensures strict alignment between the segment endpoints and the grid boundaries, while preserving the geometric details within the segment.

[0063] Flight segment collection Based on the sequential adjacency relationships of segments along the same representative route, a directed adjacency matrix of segments is constructed. ; This matrix is ​​an n×n sparse matrix, where n is the total number of flight segments, where if the flight segments... Destination and flight segment The starting point is the same dividing point or flight segment. Directly adjacent flight segments If the condition is met, the element in the i-th row and j-th column of the matrix is ​​set to 1; otherwise, it is set to 0. This directed adjacency matrix is ​​used for quick querying of the connection relationships between flight segments during subsequent network reconstruction.

[0064] This application determines the grid sequence traversed by the flight path and extracts the anchor points entering and exiting the grid. Then, it uses the perpendicular bisector of the line connecting adjacent anchor points as a segmentation baseline to cut the representative flight path. This automatically divides a complete long flight path into several discrete segments strictly aligned with the grid boundaries, ensuring that the spatial scale of each segment matches the precision of its respective grid, thus achieving multi-precision representation of the route. Simultaneously, by constructing a directed adjacency matrix for the flight segments, the sequential connection relationships between segments in the original flight path are preserved. This provides clear adjacency information for subsequent spatial and heading clustering and topological reconstruction of the navigation network, avoiding the loss of connection relationships between segmented flight paths.

[0065] In some embodiments, the step of constructing a vector navigation network by spatial and heading clustering of the target route segments specifically includes: employing a two-layer sorting point identification clustering structure algorithm to sort the spatial density of the route segments in the spatial layer, generating a reachability map based on Hausdorff distance, and identifying routes exceeding a preset steepness threshold as cluster boundaries by calculating the steepness of the similarity changes between adjacent routes in the reachability map, thereby segmenting the reachability map to obtain spatial clusters of the routes; in the heading layer, for each spatial cluster, calculating the average heading angle of the route, using the difference between the cumulative sum of the turns of two routes as a heading similarity measure, constructing a reachability map in the heading dimension, and extracting the final cluster from the reachability map using a steepness-based clustering extraction method; calculating the spatial centrality score of each route in the final cluster, selecting the route with the highest score as the representative route, extracting the feature points of the representative route as waypoints, and mapping the waypoints to routes according to the directed connectivity between the waypoints.

[0066] Specifically, at the spatial level, the OPTICS algorithm, which identifies clustering structures based on sorted points, is used to sort the samples by spatial density. For all flight segments, the Hausdorff distance between each pair of flight segments is calculated as a measure of spatial dissimilarity. A complete graph is constructed using each flight segment as a node and the Hausdorff distance as the edge weight. Then, the Prim or Dijkstra algorithm is run to generate a minimum spanning tree. The reachability distance graph is generated by sorting the edges in the minimum spanning tree by distance. The rate of change of the difference in reachability distance between adjacent flight segments in the reachability distance graph is calculated, i.e., kurtosis. A kurtosis threshold (e.g., 0.2) is set, and edges with kurtosis exceeding the threshold are identified as cluster boundaries. The reachability distance graph is then segmented to obtain several spatial clusters of flight segments, where flight segments within each cluster are spatially close to each other.

[0067] At the heading level, each spatial cluster is processed separately. The average heading angle (ranging from 0° to 360°) for each segment within that cluster is calculated. After normalizing the heading angle, the heading similarity metric between two segments is defined as the difference between their cumulative turning sums, where the cumulative turning sum refers to the sum of the absolute values ​​of the heading changes at each turning point within the segment. Based on this heading similarity metric, a reachability map along the heading dimension is reconstructed. Similarly, a kurtosis-based clustering extraction method is used to further subdivide each spatial cluster into subclusters with consistent headings, resulting in the final clusters.

[0068] For each final cluster, a spatial centrality score is calculated for each segment within it. Based on the geometric centrality score, representative waypoints and routes for each segment are extracted. This metric reflects the degree of deviation of a segment from the overall spatial distribution within the cluster by calculating a negative standardized score of the average distance between segments. For each segment within a cluster... The flight segment with the highest spatial centrality score is extracted as the representative flight segment, and the feature points of the representative flight segment are extracted as waypoints. Spatial centrality score The calculation method is as follows:

[0069] ; ; ; in, For the segment to the flight segment The distance to Hausdorff, For the segment The average distance to all other segments in the cluster, where n is the total number of segments in the current cluster. This is the mean of the average distance across all flight segments. This represents the standard deviation of the average distance across all flight segments. A higher value indicates that the segment is closer to the geometric center of the segment cluster, while a lower value indicates that the segment is closer to the edge of the cluster. Based on the directed connections between feature points in the segment, the sequence of feature point connections is mapped sequentially to routes within the cluster. Simultaneously, based on the segment adjacency matrix, the adjacency relationships of the segments are mapped to routes between the first and last waypoints of the cluster. The set of all routes constitutes the vector navigation network.

[0070] This application employs a two-layer sorted point identification clustering structure algorithm. First, density clustering of navigation segments is performed at the spatial layer based on Hausdorff distance and kurtosis. Then, at the heading layer, heading similarity clustering is performed for each spatial cluster. This approach simultaneously considers the spatial positional consistency and navigation direction consistency of navigation segments, effectively distinguishing parallel navigation segments on different waterways. Representative navigation segments are selected from each final cluster by calculating spatial centrality scores, and the feature points of these representative segments are mapped to waypoints, thus constructing a compact and geometrically representative vector navigation network. This network is directly derived from typical routes in actual AIS trajectories, accurately reflecting the habitual routes of ships at sea.

[0071] In some embodiments, the step of constructing a grid navigation network based on land boundary constraints and a triangulation method specifically includes: extracting grid edges and diagonals as grid routes based on the quadtree spatial grid model to construct an initial grid navigation network; introducing a quadtree hierarchical division strategy to perform multi-scale refinement on the initial grid navigation network, generating a concave hull boundary of the refined initial grid navigation network by adopting an alpha shape algorithm, and calculating the global water area coverage rate covered by the concave hull boundary, and if the coverage rate is lower than a preset coverage rate threshold, performing the next-level quadtree division on the uncovered water area until the coverage rate is not lower than the preset coverage rate threshold; identifying waypoints with reduced connectivity due to land boundary constraints as degraded waypoints, constructing a Delaunay triangulation network based on the degraded waypoint set, and retaining triangular edges that do not intersect with the original grid routes and land boundaries as new routes.

[0072] Specifically, grid edges and diagonals are extracted as grid routes based on the quadtree spatial grid model. A quadtree hierarchical division strategy is introduced to construct a multi-precision grid navigation network, an Alpha Shape algorithm is adopted to generate the concave hull boundary of the navigation network, and the global water area coverage rate thereof is calculated as the basis for quadtree hierarchical division. First, a global latitude and longitude grid is constructed. For each grid unit, its grid edges and diagonals are extracted, the two diagonals are mutually segmented along their intersection point, the four vertices of the grid and the diagonal intersection point are taken as grid waypoints, and bidirectional topological connecting edges between pairs of the above waypoints are constructed as grid routes. Subsequently, grid routes that cross land are eliminated, an Alpha Shape algorithm is used to obtain a concave hull polygon for the remaining routes, the proportion p of the water area covered by the polygon in the global water area is calculated, the network coverage threshold is set as p_min, and if p < p_min, the grid scale is reduced to the next level, and quadtree grid division at this scale is performed on the water area not covered by the polygon.

[0073] The discretization of grid modeling and the constraints of land boundaries limit the coverage of basic navigation networks in some coastal waters. Therefore, this application defines waypoints with reduced connectivity due to land boundary constraints as degenerate waypoints and enhances the topological connectivity of the grid navigation network using the Delaunay triangulation algorithm. A grid vertex waypoint is located at the vertex of a grid cell, and from this waypoint, there are at most eight routes, covering horizontal, vertical, and diagonal directions. A grid center waypoint is located at the geometric center of a grid cell, and from this waypoint, there are at most four routes, representing the diagonal directions of the grid cell. Therefore, the total number of routes associated with a single waypoint is defined as the connectivity of that point. Grid vertex waypoints with a connectivity of less than 8 and grid center waypoints with a connectivity of less than 4 are identified as degenerate waypoints. Based on this, a Delaunay triangulation network is constructed based on the set of degenerate waypoints, retaining the triangular edges that do not intersect with the original grid routes and land boundaries as new routes.

[0074] This application extracts the edges and diagonals of leaf grids as initial grid routes, combines them with the Alpha Shape Algorithm to calculate coverage, and dynamically refines the grid levels for uncovered waterways. This enables the automatic construction of a grid navigation network covering major global waterways while strictly avoiding land boundaries. To address the connectivity degradation caused by land boundary constraints, Delaunay triangulation is further utilized to add non-land-intersecting triangular edges between degraded waypoints as new routes, effectively improving the connectivity and integrity of the grid network in complex environments such as narrow waterways and port access areas. This method ensures that the grid navigation network can provide reasonable navigable paths even in areas without measured trajectories, complementing vector networks.

[0075] In some embodiments, the step of introducing a waypoint matching strategy based on node connectivity and spatial similarity to establish a topological association between vector and grid waypoints specifically includes: performing matching calculations on vector waypoints within a neighborhood of a grid waypoint radius τ, calculating the comprehensive matching degree between each vector waypoint and the current grid waypoint, selecting the vector waypoint with the highest comprehensive matching degree, updating the coordinates of the grid waypoint with weighted average, and synchronously updating the geometric spatial coordinates of all associated waypoints.

[0076] Specifically, for grid waypoints with a radius of... The system performs matching calculations on vector waypoints within their neighborhood, calculating the overall matching degree between each vector waypoint and the corresponding grid waypoint. The waypoint with the highest overall matching degree is selected, and its coordinates are updated using a weighted average. Simultaneously, the geometric coordinates of all associated waypoints are updated. (Waypoint overall matching degree) The calculation method is as follows:

[0077] ; ; ; in, This is the distance between two waypoints. To search for the standard deviation of the distances between all waypoints and grid waypoints within the circular search area, The connectivity of vector waypoints. and These represent the maximum and minimum connectivity of all waypoints within the search circular region, respectively. and These are the normalized scores for spatial distance and connectivity, respectively.

[0078] This application achieves precise association between actual waypoints in the vector navigation network and abstract waypoints in the grid navigation network by searching for vector waypoints within their neighborhood and calculating the matching degree by combining spatial distance similarity and topological connectivity similarity. The coordinates of successfully matched grid waypoints are updated with weighted averages, while the geometric coordinates of the associated waypoints are updated simultaneously, achieving consistent fusion of the two networks at the topological node level. This approach retains the high geometric accuracy of the vector network derived from real trajectories and inherits the spatial coverage integrity of the grid network, laying the topological and geometric foundation for the subsequent construction of a unified and usable vector-grid fused navigation network.

[0079] In some embodiments, the step of constructing a matching model based on direction, length, and geospatial similarity for arc segment matching specifically includes: taking the number of routes heading towards and out of a waypoint as the in-degree and out-degree of the current waypoint, respectively; extracting linear route sequences between waypoints with an in-degree or out-degree greater than 1 as route arc segments; performing grid encoding on the route arc segments to obtain the grid sequence corresponding to each route arc segment; extracting route arc segments with common grid subsequences as associated route arc segment pairs; calculating the direction similarity, length similarity, and geospatial similarity of the associated route arc segment pairs; fusing the direction and length geometric similarity features through weighted summation; and introducing geographical environment constraints as binary gate variables to calculate the comprehensive matching degree. If the comprehensive matching degree is not less than a preset matching degree threshold, the arc segment pair is considered a match.

[0080] Specifically, road traffic network integration can be viewed as the fusion of connecting roads between intersections. Similarly, maritime navigation network integration can be viewed as the fusion of connecting routes between waypoints. Therefore, defining the number of routes heading towards and out of a waypoint as the in-degree and out-degree of that waypoint, respectively, indicates that the waypoint has traffic flow convergence or bifurcation characteristics when the in-degree or out-degree of the waypoint is greater than 1. Linear route sequences between such waypoints are extracted as waypoint arcs. To achieve waypoint arc matching, a quadtree spatial grid model is first introduced to perform grid encoding on the waypoint arcs, i.e., counting the number of waypoint nodes in different navigation networks falling within the grid range, and using quadtree partitioning to ensure that the number of nodes in the same ship type navigation network within each grid is at most 1. Subsequently, the waypoint arc node sequence is mapped to a grid sequence, and waypoint arc combinations with common grid subsequences are extracted as associated waypoint arc pairs. The comprehensive matching degree is calculated by considering the direction, length and geospatial features of the associated route arc pairs, extracting the route arcs with matching relationships, and constructing the route arc matching matrix.

[0081] Directional similarity ( The cosine similarity is calculated by constructing the average tangent vectors of the old and new arc segments, and normalization is used to eliminate the influence of dimensional differences between different features on the similarity measurement. The specific calculation formula is as follows:

[0082] ; in, and These are the direction vectors for the old and new route arcs, respectively. and These represent the magnitudes of the directional vectors for the old and new route segments, respectively.

[0083] Length similarity ( The absolute distance error is converted into a relative distance error ratio by maximum value normalization, and then converted into a positive correlation. The specific calculation method is as follows: ; in, The cumulative length of the path is calculated from the sum of the semi-versus distances between adjacent sampling points along the path. For the flight path arc Length, For the flight path arc The length.

[0084] Geographical similarity ( While direction and length features can measure the spatial similarity of new and old arc segments in terms of geometric shape, similarity models relying solely on numerical calculations may ignore the geographical constraints of actual navigation, leading to navigation network fusion results that violate safety criteria. Therefore, a concave polygon is constructed for the route arc segments. By performing intersection operations between the concave polygon region and the land boundary, it is determined whether there are land obstacles in the waters between the arc segments, thus ensuring that the matching results of the route arc segments conform to the physical laws of actual navigation. The specific calculation formula is as follows:

[0085] ; in, The minimum concave hull polygon for the nodes of the old and new arc segments. The polygon represents a land obstacle.

[0086] In summary, the overall matching degree is calculated by weighted summation of directional and length geometric similarity features, and by introducing geographical environment constraints as a binary gating variable. A minimum overall matching degree threshold is set. If the overall matching degree between arc segments is not less than If so, then the pair of arc segments is considered a match. Overall matching degree. The calculation formula is as follows:

[0087] ; Among them, weight parameters and To balance direction and length similarity, for example, both are set to 0.5, ensuring that both are equally important for arc-path matching.

[0088] This application extracts linear sequences between nodes with an in-degree or out-degree greater than 1 in the route network as route segments and uses grid coding to quickly filter out potentially matching segment pairs, significantly reducing the computational complexity of segment matching. Based on this, it integrates directional similarity, length similarity, and geospatial overlap, and introduces geographical environmental constraints as binary gating variables to calculate the comprehensive matching degree, accurately identifying functionally equivalent or highly overlapping route segments in navigation networks of different ship types. This matching model exhibits good robustness to differences in segment geometry caused by factors such as ship size and speed, providing a reliable matching basis for subsequent multi-ship type route fusion.

[0089] In some embodiments, the steps of performing route fusion and incremental reconstruction on the vector-grid navigation network specifically include: traversing the route arc matching matrix; for fully matched arc pairs, performing weighted average fusion on the coordinates of waypoints with the same index in the arc pair; for partially matched arc pairs, extracting continuous sub-arc pairs with the same matching relationship, and performing weighted average fusion on the coordinates of waypoints with the same index in the matched sub-arc pairs; performing weighted average calculation on the coordinates of the start and end waypoints of mismatched sub-arcs respectively, so as to insert the start and end waypoints shared by the mismatched and matched sub-arcs, and merging the remaining mismatched sub-arcs with their original coordinates and topology as new routes; and merging the mismatched arc pairs with their original coordinates and topology as new routes.

[0090] Specifically, this application defines three types of route arc fusion operators to fuse route pairs under different matching scenarios in route arc pairs.

[0091] Weighted fusion operator: Without altering the number of routes or the topology of the original navigation network, it performs a weighted calculation on the spatial coordinates of waypoints to update the spatial position of the routes. Let the original route pairs be... and By and For waypoints with the same location index, a weighted average of their spatial coordinates is calculated to obtain the weighted fused waypoint. .

[0092] Insertion fusion operator: By performing coordinate weighted calculations on the starting or ending route points in a route pair, topological connections are created between route segments, thereby achieving incremental route construction and topology updates. Let the original route pair be... and The endpoints with matching relationships are extracted, and their spatial coordinates are weighted and averaged to form the common waypoints of the waypoint pairs. The remaining waypoints retain their original spatial coordinates, resulting in the fused waypoint arc pairs. and .

[0093] Merging operator: Merges route pairs into the navigation network while preserving their original topological connections and spatial coordinates. Let the original route pairs be... and If there are no routes in the navigation network and If a matching relationship exists, then... and New routes are constructed using waypoint coordinates and topological connections, and then added to the navigation network.

[0094] By traversing the route arc matching matrix Each matching relationship in ,analyze Related flight path arc and The sequence relationship between them is used to classify the associated route arc pairs into three types: arc pair complete match, arc pair partial match, and arc pair mismatch, and then perform fusion processing on each type.

[0095] Arc pair perfect matching: A weighted fusion operator is used to merge and update all routes with the same index in the arc pair. The average of their coordinates is calculated as the coordinates of the updated waypoints, resulting in a new arc composed of the updated waypoints.

[0096] Arc pair mismatch: Route arcs that do not match any other arcs are defined as isolated arcs, and the merge operator is used to directly merge isolated arcs into the original navigation network.

[0097] Arc pair partial matching: First, starting from the starting index route of the arc pair, traverse sequentially to extract consecutive sub-arc pairs with the same matching relationship. For matching sub-arc pairs, perform coordinate weighted fusion on all routes. For mismatched sub-arc pairs, if a matching sub-arc pair exists before it, insert fusion on its starting route, calculating the average coordinates of the starting route point so that the sub-arc pairs share the same starting route point; if a matching sub-arc pair exists after it, insert fusion on its ending route, calculating the average coordinates of the ending route point so that the sub-arc pairs share the same ending route point. Finally, merge and fuse the remaining unprocessed route pairs in the sub-arc pairs.

[0098] This application addresses three types of arc pair types: fully matched, partially matched, and mismatched. It employs differentiated fusion strategies, including weighted average fusion, continuous sub-arc fusion and mismatched sub-arc insertion, and directly retaining the original arc as a new route. These strategies organically integrate route information from different ship types into a unified multi-ship type fusion navigation network while maintaining the original network topology. This fusion process eliminates redundant routes caused by ship type differences while preserving the unique navigation paths of each ship type (such as differences in near-shore navigation due to different drafts). The resulting navigation network can simultaneously meet the diverse navigation needs of various ship types in terms of safety and economy.

[0099] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0100] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, database, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0101] The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention's specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for generating a maritime navigation network based on vector-grid model fusion, characterized by the following steps: include: Acquire raw AIS trajectory data of different ship types during their navigation at sea, and perform preprocessing; For each ship type, based on port stay time, berthing trajectory segments are identified from the preprocessed AIS trajectory data, continuous trajectories between adjacent ports of call are extracted as voyage trajectories, and abnormal trajectory identification and high-quality route extraction are performed on the voyage trajectories to obtain representative routes. A quadtree spatial grid model is constructed using known offshore distances and port distributions. Based on the entry and exit grid anchor points in the quadtree spatial grid model, the representative route is segmented with high precision and reconstructed into topology to obtain the target route segment. The target route segment is clustered spatially and directionally to construct a vector navigation network. A grid navigation network is constructed based on land boundary constraints and triangulation methods. A waypoint matching strategy based on node connectivity and spatial similarity is introduced to establish the topological association between vector and grid routes and generate a vector-grid navigation network. The route arcs in the vector-grid navigation network corresponding to different ship types are extracted, and a matching model based on direction, length and geospatial similarity is constructed to perform arc matching. The vector-grid navigation network is then fused and incrementally reconstructed to generate a multi-ship type fused maritime navigation network.

2. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 1, characterized in that, The steps for identifying the parking trajectory segment specifically include: The system detects trajectory points located within the same port area, identifies continuous trajectory points with a dwell time span exceeding a preset time threshold as berthing trajectory segments, and marks the port where the trajectory points are located as the port of call. Traverse the berthing trajectory segments in chronological order. If two adjacent berthing trajectory segments stop at different ports, extract the trajectory points between the end point of the previous berthing trajectory segment and the start point of the next berthing trajectory segment as a voyage trajectory segment.

3. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 1, characterized in that, The steps of identifying abnormal trajectories and extracting high-quality routes for the voyage trajectory specifically include: The voyage trajectory was subjected to cubic spline interpolation and smoothing. For the smoothed course trajectory, the penalty term is adjusted by minimizing the sum of the squared error of the heading angle and the number of feature points within the segment, and by introducing an evaluation index based on the number of feature points and Hausdorff distance to adaptively fine-tune the penalty parameters in the penalty term in order to extract the turning feature points. Meanwhile, for the smoothed voyage trajectory, a sliding window is constructed to detect the distribution of inflection points in the longitude and latitude directions of the trajectory within the window. The redundancy of the zigzag-type wandering trajectory and the redundancy of the loop-type wandering trajectory are calculated respectively. When the redundancy is greater than the preset redundancy threshold, the corresponding abnormal trajectory segments are removed. For voyage trajectories after removing abnormal trajectory segments, voyage trajectories with the same origin and destination ports are grouped together. Using Hausdorff distance as a similarity index, hierarchical clustering is adopted and a silhouette coefficient is introduced to determine the global optimal distance threshold. The trajectory with the shortest voyage in each cluster is extracted as the representative route.

4. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 3, characterized in that, The steps for constructing a quadtree spatial grid model using known offshore distances and port distribution specifically include: The global ocean area is mapped onto the six projection planes of a cube using the cube projection method. A regular grid is constructed on each projection plane to form a basic grid model. Traverse all grids, calculate the offshore scale ratio of each grid based on the nearest distance of the grid region to the coastline and the current grid side length, and count the number of ports in each grid. Within the maximum number of iterations, if the number of ports in the current grid is 0 and the offshore scale ratio is greater than the maximum offshore scale ratio, the segmentation process of the current grid ends; otherwise, quadtree segmentation is performed on the current grid, and the segmented subgrids are processed recursively.

5. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 4, characterized in that, The steps of performing multi-precision segmentation and topology reconstruction of the representative route based on the entry and exit grid anchor points in the quadtree spatial grid model specifically include: Based on the positional relationship between the route nodes and the grid, the grid sequence traversed by the route is determined, and the intersection points of the route and the boundaries of each grid are extracted as the entry and exit anchor points of the grid. Determine the vertical bisectors of adjacent entry and exit grid anchor points, and use the vertical bisectors as dividing baselines to divide the representative route into several discrete segments, wherein the two ends of the segment are dividing anchor points, and the turning feature points of the route are retained in the middle. Construct a directed adjacency matrix for each segment based on the adjacency relationships of segments along the same route.

6. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 5, characterized in that, The step of constructing a vector navigation network by spatial and heading clustering of the target route segment specifically includes: A two-layer sorting point identification clustering structure algorithm is adopted. The spatial density of the flight segments is sorted in the spatial layer to generate a reachability map based on Hausdorff distance. By calculating the steepness of the similarity change between adjacent flight segments in the reachability map, flight segments that exceed the preset steepness threshold are identified as cluster boundaries. The reachability map is then segmented to obtain the flight segment spatial clusters. At the heading layer, for each spatial cluster, the average heading angle of the flight segment is calculated. The difference between the cumulative sum of the turns of two flight segments is used as the heading similarity measure. A reachability map in the heading dimension is constructed, and the final cluster is extracted from the reachability map using a kurtosis-based clustering extraction method. Calculate the spatial centrality score of each segment in the final cluster, select the segment with the highest score as the representative segment, extract the feature points of the representative segment as waypoints, and map the waypoints as routes according to the directed connectivity between the waypoints.

7. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 6, characterized in that, The steps for constructing a grid navigation network based on land boundary constraints and triangulation methods specifically include: Based on the quadtree spatial grid model, grid edges and diagonals are extracted as grid routes to construct an initial grid navigation network. A quadtree hierarchical partitioning strategy is introduced to refine the initial grid navigation network at multiple scales. The Alpha Shape Algorithm is used to generate the concave hull boundary of the refined initial grid navigation network, and the global water area coverage rate covered by the concave hull boundary is calculated. If the coverage rate is lower than a preset coverage rate threshold, the next level of quadtree partitioning is performed on the uncovered water area until the coverage rate is not lower than the preset coverage rate threshold. For waypoints whose connectivity is reduced due to land boundary constraints, they are identified as degraded waypoints, and a Delaunay triangulation is constructed based on the set of degraded waypoints. Triangular edges that do not intersect with the original grid waypoints and land boundaries are retained as new waypoints.

8. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 7, characterized in that, The steps for establishing topological associations between vector and grid routes by introducing a waypoint matching strategy based on node connectivity and spatial similarity specifically include: For the vector waypoints within a neighborhood of radius τ of the grid waypoint, a matching calculation is performed. The comprehensive matching degree between each vector waypoint and the current grid waypoint is calculated. The vector waypoint with the highest comprehensive matching degree is selected, and the coordinates of the grid waypoint are updated with weight. The geometric space coordinates of all associated waypoints are updated simultaneously.

9. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 1, characterized in that, The steps for constructing a matching model based on direction, length, and geospatial similarity to perform arc matching specifically include: The number of routes to and from a waypoint is taken as the in-degree and out-degree of the current waypoint, respectively. The linear route sequence between waypoints with an in-degree or out-degree greater than 1 is extracted as the route arc. The route arc segments are grid-coded to obtain the grid sequence corresponding to each route arc segment. Route arc segments with common grid subsequences are combined and extracted into associated route arc segment pairs. The direction similarity, length similarity, and geospatial similarity of the associated route arc pairs are calculated. The direction and length geometric similarity features are fused by weighted summation, and geographical environmental constraints are introduced as binary gating variables to calculate the comprehensive matching degree. If the comprehensive matching degree is not less than the preset matching degree threshold, the arc pair is considered a match.

10. The method for generating a maritime navigation network based on vector-grid model fusion according to claim 9, characterized in that, The steps of performing route fusion and incremental reconstruction on the vector-grid navigation network specifically include: Traverse the route arc matching matrix. For a perfectly matched pair of arcs, perform a weighted average fusion of the coordinates of the waypoints with the same index in the arc pair. For partially matched arc pairs, extract continuous sub-arc pairs with the same matching relationship, and perform weighted average fusion on the waypoint coordinates with the same index in the matched sub-arc pairs; for the start and end waypoint coordinates of non-matching sub-arcs, perform weighted average calculation to insert the start and end waypoints shared by the non-matching and matching sub-arcs, and retain the original coordinates and topology of the remaining non-matching sub-arcs as new routes for merging and fusion. For mismatched arc pairs, the original coordinates and topological relationships are retained and merged as new routes.

Images