Ship dynamic risk group identification method and system based on spectral clustering

Abstract

The invention relates to the technical field of sea area intelligent shipping, in particular to a ship dynamic risk group identification method and system based on spectral clustering, which are used for identifying high-risk ship interaction groups and evaluating spatial distribution characteristics of the high-risk ship interaction groups. According to the method, the collision risk of each pair of ships is calculated in real time by obtaining AIS dynamic data, and a risk weight matrix is constructed by indexes; introducing modularity as a ship clustering cluster number estimation standard, and realizing ship group division based on a spectral clustering algorithm; further combining modularity optimization and risk potential field calculation, and determining and identifying the positions of a dynamic high-risk group and a water area hot spot area; according to the method, potential conflict groups can be dynamically extracted in a complex navigation environment, accurate modeling of risk structures among ships is achieved, intelligent and prospective support is provided for maritime affair supervision and risk intervention, and therefore ship navigation safety is guaranteed.

Description

Technical Field

[0001] This application relates to the technical field of intelligent shipping in sea areas, and particularly to a method and system for identifying ship dynamic risk groups based on spectral clustering. Background Art

[0002] With the continuous growth of shipping volume, the traffic density in coastal waters has increased, and the navigation environment has become more complex, which further leads to a significant increase in the probability of ship navigation accidents. Currently, common ship risk assessment methods include static collision parameter analysis methods based on DCPA (distance to closest point of approach) and TCPA (time to closest point of approach), intelligent modeling methods based on fuzzy logic, artificial potential field method, etc., and rule-based risk index calculation models (such as Collision Risk Index, CRI). These methods have certain effects in analyzing the risk relationship between single pairs of ships, but in the face of large-scale multi-ship cooperation, intensive intersections, and group behavior aggregation situations, they lack in-depth modeling of the risk evolution trend between groups and are difficult to meet the global risk identification and intervention requirements in the modern navigation environment.

[0003] Especially in traffic-intensive areas such as coastal waters, near-port areas, or channel convergence areas, situations where multiple ships meet, cross, or gather and wait simultaneously are likely to form potential high-risk groups, and existing research often ignores the impact of the complex interaction relationship between ships on the group structure stability and systematic risk conduction path. In addition, most existing clustering methods (such as DBSCAN, K-means, etc.) define clustering boundaries based on Euclidean space density or distance, and fail to effectively introduce the graph theory modeling idea driven by risk indicators, resulting in unclear group division boundaries and ineffective identification of interfering ships, which limits the forward-looking and accuracy of risk assessment.

[0004] In view of this, the present invention proposes a method and system for identifying ship dynamic risk groups based on spectral clustering, which can achieve accurate identification of the ship group structure, thereby providing intelligent support for navigation safety assessment, traffic guidance, and ship cooperative collision avoidance. Summary of the Invention

[0005] In order to solve the problems in the prior art such as unclear group division boundaries and ineffective identification of interfering ships, the present invention provides a method and system for identifying ship dynamic risk groups based on spectral clustering to solve the above technical defect problems.

[0006] In a first aspect, the present invention proposes a method for identifying ship dynamic risk groups based on spectral clustering, the method comprising the following steps: S1. Collect AIS data in coastal waters, decode, clean, repair, and interpolate the AIS data, and divide the processed data by time window to obtain preprocessed data; S2. Based on the preprocessed data, construct a ship encounter situation, and calculate the collision parameters between the own ship and the target ship. The collision parameters include: the relative distance between the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to closest point of approach (TCPA) between the own ship and the target ship, and the closest distance of approach (DCPA) between the own ship and the target ship; S3. According to the ship encounter situation, assign different weights to the shortest time TCPA and the shortest distance DCPA, and calculate the collision risk value between ship pairs; S4. Taking all ships as nodes and the risk values between ship pairs as the weights of the edges, construct a ship risk connection relationship network graph; S5. Based on the risk adjacency matrix of the ship risk connection relationship network graph, calculate the standard Laplacian matrix, and extract multiple eigenvectors corresponding to the smallest eigenvalue of the standard Laplacian matrix, and arrange them column by column to form an eigenmatrix; Use the K-means clustering algorithm to perform clustering analysis on the eigenmatrix to obtain the ship dynamic risk group division result.

[0007] Preferably, in step S3, according to the ship encounter situation, assign different weights to the shortest time TCPA and the shortest distance DCPA, and calculate the collision risk value between ship pairs, which specifically includes the following sub-steps: S31. Establish a relationship formula between the shortest time TCPA, the shortest distance DCPA and the risk value by using the negative exponential function model to calculate the collision risk value of the ship in space and time. The specific calculation formula is:

[0008] In the formula, represents the collision risk value based on the shortest distance DCPA; represents the collision risk value based on the shortest time TCPA; S32. According to different ship encounter situations, adopt different weight assignment strategies for to calculate the collision risk value between ship and ship . The calculation expression is:

[0009] In the formula, represents the comprehensive collision risk value between ship and ship ; represents the weight coefficient in a specific ship encounter situation, and its value range is (0, 1); represents the weight complement corresponding to the shortest time TCPA; k represents the type number of the ship encounter situation, and its value is 1, 2 or 3, representing overtaking, head-on encounter, and crossing encounter respectively.

[0010] Preferably, in step S4, a ship risk connection relationship network diagram is constructed with all ships as nodes and the risk values between ship pairs as the weights of the edges, which specifically includes the following sub-steps: S41. Abstract all ships in the coastal waters as nodes of the relationship network diagram; S42. According to the collision risk values between ship pairs calculated in step S3, establish connection edges between nodes, where: If the distance to closest point of approach (DCPA) and the time to closest point of approach (TCPA) between ship and ship are both lower than the preset safety threshold, then establish a connection edge; If any one of the parameters of the DCPA or TCPA exceeds the preset safety threshold, it is regarded as having no risk interaction and no connection edge is established; S43. Construct an undirected weighted graph with the collision risk value as the weight of the edge;

[0011] Preferably, in step S5, before calculating the standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection relationship network diagram, it also includes calculating the modularity of the ship risk connection relationship network diagram to determine the optimal number of clustering clusters, which specifically includes the following sub-steps: S511. Calculate the degree of node and the degree of node ; S512. Calculate the total edge weight value m in the ship risk connection relationship network diagram; S513. Calculate the expected edge weight value between node and node in the random network, S514. According to the difference between the actual edge weight value and the expected edge weight value , calculate the obtained modularity M, and the calculation expression is:

[0012] In the formula, , indicating that node and node belong to the same group, otherwise it is 0; S515. Traverse different numbers of clustering clusters, and select the number of clusters that maximizes the modularity M as the number of ship groups adopted in spectral clustering.

[0013] Preferably, in step S5, based on the risk adjacency matrix of the ship risk connection relationship network diagram, a standard symmetric normalized Laplacian matrix is constructed, and its calculation expression is:

[0014] In the formula, I is the identity vector; D is the node degree matrix, which represents the number of nodes in the ship risk connection relationship network and is an n-order diagonal matrix, the diagonal elements are the degree values of the corresponding nodes, the non-diagonal elements are all 0, and n is the number of ships; W is the risk adjacency matrix, and the risk adjacency matrix W is a symmetric matrix corresponding to an undirected graph, its diagonal elements are all 0, and the non-diagonal elements are 0 or 1.

[0015] Preferably, in step S5, it also includes posterior risk assessment of the clustered ship dynamic risk groups, which specifically includes the following sub-steps: S521. Let the target ship in a certain group be P, and traverse the collision risk values between other ships in the group and ship P , and classify according to the following rules: If the collision risk value ∈[0, 0.4], then classify the risk relationship as type I data; If the collision risk value ∈(0.4, 0.8), then classify the risk relationship as type II data; If the collision risk value ∈[0.8, 1], then classify the risk relationship as type III data; S522. According to the data categories divided in step S521, calculate the comprehensive collision risk value CCRV of each ship: For type I data, the comprehensive collision risk value CCRV is 0; For type II data, the comprehensive collision risk value CCRV is calculated through the following formula:

[0016] In the formula, is the comprehensive collision risk value of type II data; m is the number of type II data; For type III data, the comprehensive collision risk value CCRV is calculated through the following formula:

[0017] In the formula, is the comprehensive collision risk value of type III data; n is the number of type III data; S523. Calculate the total risk value TCR of the group according to the comprehensive collision risk value CCRV, and determine the maximum single-ship risk value MCR of the group. Obtain the first collision risk degree value according to the total risk value TCR of the group , obtain the second collision risk degree value according to the maximum single - ship risk value MCR , and the expression is as follows:

[0018] In the formula, is the first collision risk degree value of group A; is the total risk value of group A; is the maximum value among the total risk values TCR of all groups; is the second collision risk degree value of group A; is the maximum single - ship risk value of group A; is the maximum single - ship risk value among all groups; S524. Determine the larger value between the first collision risk degree value and the second collision risk degree value as the final collision risk degree of the water area where group A is located.

[0019] Preferably, in step S5, the K - means clustering algorithm is used to perform clustering analysis on the feature matrix, which specifically includes the following sub - steps: S531. Perform eigenvalue decomposition on the standard Laplacian matrix, and extract the eigenvectors corresponding to the first smallest non - zero eigenvalues; S532. Arrange the eigenvectors by column to form a - dimensional feature matrix Q, where is the number of ship nodes, is the optimal number of clustering clusters; S533. Perform standardization processing on each row of the feature matrix Q to obtain the standardized feature space data; S534. Use the K - means algorithm to perform clustering on the feature space data, divide the ship dynamic risk groups, and output the group labels of each ship node.

[0020] Preferably, in step S2, the calculation formulas for the shortest time to CPA TCPA between the own ship and the target ship and the shortest distance DCPA between the own ship and the target ship are:

[0021] In the formula, OS and TS respectively represent the own ship and the target ship, d represents the relative distance, and respectively represent the speed of the own ship and the speed of the target ship, represents the relative speed, and β represent the relative bearing angle; The relative distance d is calculated based on the longitude and latitude of the ship through the Haversine formula. The Haversine calculation formula is as follows:

[0022] In the formula, represents a parameter; c represents the great circle arc length; represents the latitude difference, represents the latitude; represents the longitude difference, and R represents the radius of the earth; Relative speed is calculated based on the speed of the own ship and the speed of the target ship. The calculation formula for the relative speed is as follows:

[0023] In the formula, represents the course of the own ship, represents the course of the target ship.

[0024] Preferably, in step S1, the AIS data is decoded, cleaned, repaired and interpolated, and the processed data is divided according to a time window to obtain preprocessed data, which specifically includes the following sub-steps: S11. Decode the AIS data and extract the position variables, speed, route, ship attributes and environmental parameters of the ship; S22. Clear abnormal data, where the abnormal data includes longitude and latitude drift points, out-of-range data and incorrect MMSI identification codes; S23. Repair the missing trajectory data through the cubic spline interpolation method to generate a continuous trajectory sequence; S24. Divide the data at a preset time interval, construct a time window for dynamic risk analysis, and obtain preprocessed data.

[0025] In a second aspect, the present invention proposes a ship dynamic risk group identification system based on spectral clustering. The system includes: A data acquisition module configured to collect AIS data in coastal waters, decode, clean, repair and interpolate the AIS data, and divide the processed data according to a time window to obtain preprocessed data; A data processing module configured to construct a ship encounter situation based on the preprocessed data and calculate the collision parameters between the own ship and the target ship. The collision parameters include: the relative distance between the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the shortest distance DCPA between the own ship and the target ship; A risk value calculation module, configured to allocate different weights to the shortest time to CPA (TCPA) and the shortest distance to CPA (DCPA) according to the ship encounter situation, and calculate the collision risk value between ship pairs; A risk connection relationship graph construction module, configured to construct a ship risk connection network graph with all ships as nodes and the risk values between ship pairs as the edge weights; A result output module, configured to calculate a standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection network graph, and extract multiple eigenvectors corresponding to the smallest eigenvalue of the Laplacian matrix, and arrange them column by column to form a feature matrix; Use the K-means clustering algorithm to perform clustering analysis on the feature matrix to obtain the division result of the ship dynamic risk groups.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Accurately identify the ship group structure: By integrating multi-source data such as AIS dynamic information, collision parameters, encounter types, and ship behavior patterns, establish a ship conflict relationship graph with risk drive as the core, and use the spectral clustering method to divide the group structure in real time, dynamically identify high-risk navigation situations, and achieve accurate identification of the ship group structure.

[0027] (2) Effectively model the ship interaction relationship: Use collision parameters to construct the risk connection strength between ships, dynamically model the interaction relationship between ships, and combine the actual encounter scenario and expert experience rules to convert the crew's perception of risk into a computable risk index.

[0028] (3) Clearly divide risk groups: Use graph theory methods to construct a ship risk relationship network, and use the modularity function and spectral clustering method to aggregate and divide the ship group structure, and extract risk groups with clear structures.

[0029] (4) Intelligently identify and evaluate risks: Based on the risk level classification mechanism and the comprehensive collision risk value (CCRV) calculation method, perform posterior summary and quantitative evaluation on the ship risk levels in different collision scenarios, and achieve intelligent identification and evolutionary analysis of potential conflict groups.

[0030] (5) Improve navigation safety: Provide intelligent support for navigation safety assessment, traffic guidance, and ship cooperative collision avoidance, help identify risks in advance and take preventive measures, thereby improving the navigation safety of ships. Description of the Drawings

[0031] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes, and advantages of the present application will become more obvious: Figure 1It is the flowchart of the ship dynamic risk group identification method based on spectral clustering of the present invention; Figure 2 It is the schematic diagram of typical ship encounter situations in the embodiments of the present invention; Figure 3 It is the schematic diagram of the calculation principle of ship collision parameters (DCPA and TCPA) in the embodiments of the present invention; Figure 4 It is the construction diagram of the collision risk value index in the embodiments of the present invention; Figure 5 It is the change diagram of the index M of the largest connected component when selecting the number of groups at four moments in the embodiments of the present invention; Figure 6 It is the schematic diagram of the ship risk connection relationship network in the embodiments of the present invention. Detailed implementation manners

[0032] The following further elaborates the present application in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. Additionally, it should be noted that for the sake of description, only parts related to the relevant invention are shown in the drawings.

[0033] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The following will elaborate on the present application in detail with reference to the drawings and embodiments.

[0034] The present invention proposes a ship dynamic risk group identification method based on spectral clustering, Figure 1 shows the flowchart of the ship dynamic risk group identification method based on spectral clustering according to the present invention, as Figure 1 shown, the method includes the following steps: S1. Collect AIS data (Automatic Identification System data, which is the real-time navigation information broadcast by ships through radio signals) in coastal waters, decode, clean, repair, and interpolate the AIS data, and divide the processed data according to time windows to obtain preprocessed data.

[0035] In this embodiment, it specifically includes the following sub-steps: S11. Decode the AIS data and extract the position variables, speeds, routes, ship attributes, and environmental parameters of the ships. Among them, the environmental parameters include: the waters where the ships are located, weather conditions, ship visibility, and real-time ship traffic factors.

[0036] S22. Clear abnormal data, where the abnormal data includes longitude and latitude drift points, out-of-range data, and incorrect MMSI identification codes; S23. Repair the missing trajectory data through cubic spline interpolation method to generate a continuous trajectory sequence; S24. Divide the data at a preset time interval, construct a time window for dynamic risk analysis, and obtain preprocessed data.

[0037] Through the above processing, the continuity and accuracy of the trajectory can be ensured. The processed data is divided according to the time window for subsequent risk modeling.

[0038] Continue to refer to Figure 1 , a method for identifying ship dynamic risk groups based on spectral clustering proposed by the present invention further includes the following steps: S2. Based on the preprocessed data, construct a ship encounter situation, and calculate the collision parameters between the own ship and the target ship. The collision parameters include: the relative distance between the position variable of the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the shortest distance to CPA (DCPA) between the own ship and the target ship.

[0039] Among them, to construct a ship encounter situation, the method for judging the ship encounter situation is as follows: When two ships meet on opposite or nearly opposite headings (the relative azimuth angle is between 352.5° and 7.5°) and pose a collision risk, it is a head-on situation. Both the own ship and the target ship should turn right to ensure that they pass by each other's port sides; when one ship is overtaken by an approaching ship at an azimuth angle of 112.5° to 247.5°, the two ships are in an overtaking situation. When the approaching ship is within the range of 7.5° to 112.5° or 247.5° to 352.5° of the horizontal light irradiation of the own ship's side light and poses a collision threat, it is a crossing situation.

[0040] In step S2, the calculation formulas for the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship and the shortest distance to CPA (DCPA) between the own ship and the target ship are:

[0041] In the formula, OS and TS respectively represent the own ship and the target ship, d represents the relative distance, and respectively represent the speed of the own ship and the speed of the target ship, represents the relative speed, and β represent the relative course angle; The relative distance d is calculated based on the ship's longitude and latitude through the Haversine formula. The Haversine calculation formula is:

[0042] In the formula, represents the parameter; c represents the great circle arc length; represents the latitude difference, represents latitude; represents the longitude difference, and R represents the radius of the earth; Relative speed is obtained by calculating the speed of the own ship and the speed of the target ship. The relative speed The calculation formula is:

[0043] In the formula, represents the course of the own ship, represents the course of the target ship.

[0044] Continue to refer to Figure 1 , a method for identifying ship dynamic risk groups based on spectral clustering proposed by the present invention, further includes the following steps: S3. According to the ship encounter situation, different weights are assigned to the shortest time TCPA and the shortest distance DCPA, and the collision risk value between ship pairs is calculated.

[0045] In this embodiment, by setting the collision risk value, an index (Collision Risk Index, CRI) for evaluating the collision risk level of two ships during navigation is used. The value range of CRI is 0-1. When it is 0, it means that there is no collision danger between ships; when its value is 1, it means that the collision cannot be avoided no matter how the operation is carried out; when CRI is between 0-1, it means that there is a collision danger, and the closer the value is to 1, the higher the collision danger level faced by the ship.

[0046] According to the ship encounter situation, different weights are assigned to the shortest time TCPA and the shortest distance DCPA, and the risk value between ship pairs is calculated, which specifically includes the following sub-steps: S31. Use the negative exponential function model to establish the relationship between the shortest time TCPA, the shortest distance DCPA and the risk value to calculate the collision risk value of the ship in space and time. The specific calculation formula is:

[0047] In the formula, represents the collision risk value based on the shortest distance DCPA, reflecting the impact of the physical proximity between ships on the risk; represents the collision risk value based on the shortest time TCPA, reflecting the urgency of the potential collision; S32. According to different ship encounter situations, for , adopt different weight assignment strategies to calculate the collision risk value between ship and ship The calculation expression is:

[0048] In the formula, represents the comprehensive collision risk value between and ships; represents the weight coefficient under a specific ship encounter situation, and its value range is (0, 1); represents the weight complement corresponding to the shortest time TCPA; k represents the type number of the ship encounter situation, and its value is 1, 2, or 3, representing overtaking, head-on encounter, and crossing encounter respectively.

[0049] Continue to refer to Figure 1 , a method for identifying ship dynamic risk groups based on spectral clustering proposed by the present invention further includes the following steps: S4. Construct a ship risk connection relationship network graph with all ships as nodes and the risk value between ship pairs as the edge weight.

[0050] In this embodiment, it specifically includes the following sub-steps: S41. Abstract all ships in the coastal waters as nodes of the relationship network graph; S42. According to the collision risk value between ship pairs calculated in step S3, establish connection edges between nodes, where: If the shortest distance DCPA and the shortest encounter time TCPA between ship and ship are both lower than the preset safety threshold, establish a connection edge; If any one of the shortest distance DCPA or the shortest encounter time TCPA exceeds the preset safety threshold, it is regarded as a risk-free interaction and no connection edge is established; S43. Construct an undirected weighted graph with the collision risk value as the edge weight;

[0051] Continue to refer to Figure 1 , a method for identifying ship dynamic risk groups based on spectral clustering proposed by the present invention further includes the following steps: S5. Based on the risk adjacency matrix of the ship risk connection relationship network graph, calculate the standard Laplacian matrix, and extract multiple eigenvectors corresponding to the smallest eigenvalue of the standard Laplacian matrix, and arrange them in columns to form a feature matrix; Use the K-means clustering algorithm to perform clustering analysis on the feature matrix to obtain the ship dynamic risk group division result.

[0052] In this embodiment, before calculating the standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection relationship network diagram, it further includes calculating the modularity of the ship risk connection relationship network diagram to determine the optimal number of clustering clusters, specifically including the following sub-steps: S511. Calculate the degree of node and the degree of node ; S512. Calculate the total edge weight value m in the ship risk connection relationship network diagram; S513. Calculate the expected edge weight value of node and node in the random network, S514. According to the difference between the actual edge weight value and the expected edge weight value , calculate and obtain the modularity M, and the calculation expression is:

[0053] In the formula, , indicates that node and node belong to the same group, otherwise it is 0; S515. Traverse different numbers of clustering clusters, and select the number of clusters that maximizes the modularity M as the number of ship groups used in spectral clustering.

[0054] Based on the risk adjacency matrix of the ship risk connection relationship network diagram, construct a standard symmetric normalized Laplacian matrix, and its calculation expression is:

[0055] In the formula, I is the unit vector; D is the node degree matrix, which represents the number of nodes in the ship risk connection relationship network and is an n-order diagonal matrix, the diagonal elements are the degree values of the corresponding nodes, and the non-diagonal elements are all 0, and n is the number of ships; W is the risk adjacency matrix, and the risk adjacency matrix W is a symmetric matrix corresponding to an undirected graph, its diagonal elements are all 0, and the non-diagonal elements are 0 or 1.

[0056] Use the K-means clustering algorithm to perform clustering analysis on the feature matrix, specifically including the following sub-steps: S531. Perform eigenvalue decomposition on the standard Laplacian matrix, and extract the eigenvectors corresponding to the first smallest non-zero eigenvalues; S532. Arrange the eigenvectors in columns to form a dimensional feature matrix Q, where is the number of ship nodes, is the optimal number of clustering clusters; S533. Standardize each row of the feature matrix Q to obtain the standardized feature space data; S534. Use the K-means algorithm to cluster the feature space data, divide the ship dynamic risk groups, output the group labels of each ship node, and realize the group recognition structure with high risk density within the group and low coupling between groups.

[0057] The present invention fuses multi-source data such as AIS dynamic information, collision parameters, encounter types, and ship behavior patterns, establishes a ship conflict relationship graph with risk drive as the core, uses the spectral clustering method to divide the group structure in real time, and dynamically identifies high-risk navigation situations; constructs the risk connection strength between ships through collision parameters for dynamically modeling the interaction relationship between ships, and then introduces expert experience rules based on the actual encounter scenario to convert the crew's perception of the risk structure into a computable risk index; constructs a ship risk relationship network by combining graph theory methods, and uses the modularity function and spectral clustering method to aggregate and divide the ship group structure to realize the extraction of risk groups with clear structures; finally, through the risk level classification mechanism and the comprehensive risk value (CCRV) calculation method, posterior summary and quantitative evaluation of the ship risk levels in various collision scenarios are carried out to realize the intelligent identification and evolution analysis of potential conflict groups in the ship traffic system.

[0058] In a specific embodiment, the ship traffic AIS data includes the position variables, speeds, routes, waters where the ships are located, weather conditions, ship visibility, ship real-time traffic factors, and ship information of the ships, which can reflect the motion state of the ship trajectories and is conducive to establishing a ship attribute model in risk assessment and assisting in depicting risk background factors. After decoding the AIS original telegram, preprocessing such as cleaning, duplicate removal, and interpolation is performed. Using the processed AIS data, with a 5-minute interval, select the ships at four moments within 15 minutes as the research objects, and the following is a specific description in combination with the actual sea area traffic scenario.

[0059] By studying the water area range of a certain port, the types of ships sailing there are diverse, including large container ships, bulk carriers, oil tankers, passenger ferries, and fishing boats, and the routes are intertwined frequently, forming a high-density ship encounter situation. Therefore, structuring the modeling and group identification of the collision risks between ships is helpful for the decision-making of the intelligent collision avoidance system. The present invention innovatively uses the spectral clustering-based dynamic group identification method for ship collision risks to deeply analyze the ship traffic conditions in the selected scenario.

[0060] Observe the navigation status of ships in the sea area on the day of the selected scene in the port and its adjacent sea area from 11:05 to 11:20 on December 15, 2020. It is drawn at 5-minute intervals and consists of four time slices in total. It can be observed that ships inside and outside the port area are in a dense meeting situation, including anchored ships, departing ships and ships entering from the outer port. Ships with a connection relationship tend to approach each other. Among them, according to the ship AIS data, a ship encounter scenario is simulated and constructed. Using the processed AIS data, at 5-minute intervals, four moments within 15 minutes are randomly selected as the research objects, such as between 11:05 and 11:20 on December 15, 2020. Determine whether the current ship encounter scenario belongs to the following scenarios to filter the AIS data, and retain and determine three encounter scenario situations.

[0061] In this embodiment, the collision parameters between the own ship and the target ship are calculated. The collision parameters are used to describe the dynamic motion characteristics of the ship. The collision parameters include: the relative distance between the position variable of the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the calculation formulas for the shortest distance DCPA between the own ship and the target ship are as follows:

[0062] Among them, OS and TS represent the own ship and the target ship respectively, d represents the relative distance, and represent the speed of the own ship and the speed of the target ship respectively, and the relative speed V is calculated through the speed of the own ship and the speed of the target ship r , and represent the relative angle of the bow respectively.

[0063] In this embodiment, according to the ship risk data, a ship encounter scenario is extracted and constructed. The ship encounter scenario includes multiple scenario elements. Previous studies have established a collision model based on the "International Regulations for Preventing Collisions at Sea" (COLREG), and roughly divided the encounter scenarios into three types: head-on encounter, crossing encounter and overtaking. One of the limitations of previous studies is that different encounter scenarios have different avoidance responsibility judgment and collision parameter weight distribution logics. Previous studies are mostly based on fixed thresholds or single-direction models, and geometrically model the approaching process between ships from a physical perspective, ignoring the role of ship maneuverability, environmental conditions and actual traffic behavior in risk formation. For example, in the case of poor visibility, on this basis, the present invention further considers the environmental dynamics and experience rule differences in the actual navigation process, constructs a collision risk assessment mechanism with behavior response ability, can significantly improve the accuracy of risk perception, and provides a reliable basis for subsequent group identification and risk level judgment.

[0064] Furthermore, as an executable solution, a fixed ratio has often been used in previous studies to allocate the two. However, when a ship faces different navigation scenarios, its considerations for DCPA and TCPA will also vary. For example, when determining the collision risk in the case of crossing and overtaking, the proportion of DCPA in the former is higher than that in the latter. Therefore, when calculating the collision risk by comprehensively considering DCPA / TCPA in the present invention, different weight ratios are given to DCPA / TCPA according to different encounter situations of ships. First, the collision risks of DCPA and TCPA in space and time are described by an exponential function expression with unknown coefficients, and then the specific coefficient values are calculated in combination with the opinions of maritime experts. The collision risk values of DCPA and TCPA in space and time and are expressed as follows:

[0065] Since the safety threshold of DCPA is determined by referring to the ship domain when detecting the ship collision risk, the safety threshold is affected by the relative bearing of the approaching ship. To eliminate this influence in the quantitative calculation, the calculation formula needs to be processed as follows:

[0066] where a and b are the parameters for quantifying DCPA / TCPA in the collision risk. Their collision risk index values are as shown in Figure 4 .

[0067] Among them, is calculated as follows:

[0068] In this embodiment, the collision risk value is an index (Collision Risk Index, CRI) used to evaluate the collision risk level between two ships during navigation. The value range of CRI is 0 - 1. When it is 0, it means there is no collision danger between ships; when its value is 1, it means that no matter how to operate, a collision cannot be avoided; when CRI is between 0 and 1, it means there is a collision danger. The closer the value is to 1, the higher the collision danger level faced by the ship. According to the encounter situation of the ships described above, different weights are assigned to the collision parameters DCPA and TCPA. The 1st, 2nd, and 3rd encounter scenarios represent overtaking, head-on encounter, and crossing encounter respectively, and the risk value calculation between ship and ship is carried out. The calculation formula is as follows:

[0069] Among them, It corresponds to the encounter situation, and k takes values of 1, 2, and 3, corresponding to head-on situation, overtaking situation, and crossing situation of ships respectively; In order to enhance the adaptability of the model in different encounter situations, the present invention designs a scoring table according to the relevant content of the responsibility division in different encounter situations in COLREGs. For each type of encounter situation, the sensitivity of DCPA and TCPA is scored by the method of expert scoring. The scores of each scenario are summarized, and the average score of each parameter is calculated. After normalization, , , The values of are 0.542, 0.639, and 0.591 respectively; Figure 2 Figure shows the classification diagram of typical ship encounter situations in the present invention. As Figure 2 shown, a 360° azimuth system is constructed with the own ship OS as the reference point, and its front is set as the 0° / 360° direction, and the azimuth angle is measured counterclockwise. According to the International Regulations for Preventing Collisions at Sea and navigation practice, three typical encounter scenarios are divided: 1) Head-on situation: When the target ship is within the range of ±7.5° directly ahead of the OS (i.e., between 352.5° and 7.5°), the two ships are approaching each other, and both should keep clear to starboard; 2) Crossing situation: When the target ship approaches from the starboard direction of the OS (7.5° - 112.5°), it has the obligation to keep clear of the OS; similarly, when the target ship approaches from the port side direction (247.5° - 352.5°), the OS should keep clear; 3) Overtaking situation: When the target ship is in the range of the tail of the OS (112.5° - 247.5°), it is in an overtaking situation, and the overtaking ship shall be responsible for keeping clear.

[0070] Figure 3 Figure shows the schematic diagram of ship collision parameter calculation in the present invention. As Figure 3 shown, based on the inertial two-dimensional rectangular coordinate system, the relative motion relationship between the own ship and the target ship is shown, and the geometric calculation process of core collision indicators such as DCPA and TCPA is formally represented. Among them, d represents the relative distance between the two ships, and respectively represent the navigation speed vectors of the own ship and the target ship, is the relative speed vector, represents the azimuth angle of the target ship relative to the own ship, is the angle between the relative speed direction and the relative position direction.

[0071] Figure 4This is the construction diagram of the collision risk degree index in the present invention. The adopted negative exponential function model establishes the relationship between DCPA, TCPA and the risk value to calculate the collision risk degree of ships in space and time. It fully considers the non-linear relationship between collision parameters and risks, avoiding the problem of insufficient accuracy shown by traditional linear normalization methods in practical applications. Figure 4 respectively give and of the exponential function curves, showing their contribution trends to the overall collision risk in different numerical intervals. Subsequently, according to different encounter situations, different weights are assigned to and , and finally the weighted results are fused to obtain the CRI index used for actual collision risk assessment in the figure.

[0072] In order to improve the structural rationality and result stability of ship group division in the spectral clustering process, the present invention introduces the modularity function before clustering to evaluate the graph structure under different numbers of clustering clusters k. As a quantitative index for evaluating the quality of network division, modularity measures the degree of aggregation of risk connections within the group by comparing the sum of edge weights within the group in the actual risk graph with the expected number of edges in a random network. The value range of modularity is [-1, 1], and the higher its value indicates stronger risk coupling within the group, weaker risk interference between groups, and a better division structure. The present invention selects the optimal number of clusters based on the modularity function , which can avoid the subjective setting of the number of clusters in traditional methods, improve the self-adaptability of spectral clustering division, the physical interpretability of the clustering structure, and the engineering applicability of ship risk group identification. The calculation steps of modularity are as follows: 1) Calculate the degree of each node and ; 2) Calculate the total number of edges m in the network; 3) Calculate the expected number of edges: In a random network, the connection probability between each pair of nodes and is proportional to their degrees, and the expected number of edges is , is the sum of all possible numbers of edges; 4) Calculate the actual number of edges; 5) Summarize the difference between the actual number of edges and the expected number of edges to obtain the contribution of each ship, and standardize it to obtain the modularity M, and its expression is:

[0073] Among them, is the edge weight, , are the node degrees, that is, node , node The sum of the edge weights connected to all other nodes , indicating the node and the node belong to the same cluster, otherwise it is 0; after the ship group division in the present invention, the modularity index is used to evaluate the division result, and thus the number of group divisions can be determined.

[0074] Figure 5 is the dynamic change characteristic of the maximum connected component M index when selecting the number of groups at different time nodes. This is statistically calculated after the clustering numbers of the remaining groups are fixed. It can be observed that when 10, 13, 11, and 11 groups are taken for the four monitoring points respectively, the M index all shows a relatively high value range, which illustrates the effectiveness of the group number selection strategy and the accuracy of the division result. In addition, the clustering number shows a dynamic fluctuation characteristic during this period, which coincides with the spatio-temporal heterogeneity of ship behavior in the water area, and further verifies the feasibility of the dynamic group division method in the present invention.

[0075] In order to improve the ability to retain the structural characteristics of the ship conflict relationship network and the stability of clustering division in the spectral clustering process, after constructing the ship risk connection graph G in the present invention, the Laplace matrix L is constructed based on the adjacency matrix W, and its calculation method is as follows:

[0076] Among them, I is the unit vector; D is the node degree matrix. The degree matrix D is used to describe the number of nodes in the ship risk connection relationship network, and it is an n-order diagonal matrix. All elements except the diagonal elements are 0, where n is the number of nodes, that is, the number of ships; the diagonal element is the degree of the node (the number of edges connected to the node ), that is, the number of risk connection relationships with the ship . The calculation formula is: ; for the adjacency matrix W of an undirected graph, it is a symmetric matrix, and the elements on the diagonal are all 0, and the elements on the non-diagonal are all 0 or 1. When , there is no risk connection relationship between the ship and the ship ; otherwise, there is a risk connection relationship between the ship and the ship ; the Laplace matrix of the network G contains important information such as the risk connection structure and risk magnitude between ships. By analyzing the characteristics of the Laplace matrix, the characteristics of G can be obtained.

[0077] Figure 6The figure shows a schematic diagram of the ship risk connection relationship network constructed in the embodiment of the present invention. As shown in the figure, each in - service ship in the target water area is represented by a node, and the edge represents the connection relationship with a significant collision risk between ships. By calculating the collision parameters between ships in the AIS dynamic data and setting a risk threshold based on the collision risk degree index, a weighted adjacency matrix W is constructed for all ship pairs, and an undirected weighted graph G is generated accordingly.

[0078] Both the adjacency matrix W and the degree matrix D are symmetric matrices. Therefore, the Laplace matrix is also a symmetric matrix. For the Laplace matrix, for any vector and its transposed vector , the following properties exist:

[0079] Here, is the weight of the connecting edge, that is, the magnitude of the risk value between ship and . The above formula represents the total variation of the graph, which is the weighted sum of the risk value differences between all ship nodes here; fi and fj are the signal values of the graph signal at nodes and , that is, the eigenvalue of this node, which is actually the comprehensive score of the ship node in the graph in this application. When performing group division, it is hoped that the difference in node scores is as small as possible. The minimization of the above formula represents the smoothing process of the overall risk. It is hoped that adjacent ships have a small risk difference, so as to form a more balanced risk network. To identify ship groups with higher mutual risks, so that the risk differences within these groups are small, but the overall risk is high.

[0080] For the non - symmetric Laplace matrix L, a normalization process is required, and the formula is:

[0081] Perform eigenvalue decomposition on the Laplace matrix to obtain eigenvalues and eigenvectors .

[0082] The purpose of clustering is to group ships with high risk similarity into the same group, so that the sum of the number and weight of the connecting edges between groups is as small as possible, thereby obtaining the optimal division result. Suppose the ships in G are to be divided into two groups, and the number of connecting edges between the two groups is H, as follows:

[0083] Let s be an index vector containing n elements, which is used to define the group to which the ship belongs. The elements are defined as:

[0084] Then there is:

[0085] At this time, the following relationship exists between H and the node degree:

[0086] Among them, is 1, is an element in the introduced matrix. For there is:

[0087] For this, the expression of H can be changed to:

[0088] According to the properties of the Laplace matrix, H is also transformed into matrix form:

[0089] Among them , L ij The value of is:

[0090] To make H take the minimum value, let

[0091] Then , Among them, , are the eigenvalues of matrix L and their corresponding eigenvectors. When the elements in the eigenvector corresponding to the second smallest eigenvalue, i.e., the Fiedler vector, are greater than or equal to 0, the element corresponds to +1 in the index vector, otherwise it is -1. Based on this, the purpose of dividing the ships into two groups is achieved.

[0092] Matrix L is a positive semi - definite matrix, and the smallest eigenvalue is 0, that is . When the graph consists of k connected sub - graphs, holds on each connected sub - graph, and through a certain node sorting, the Laplace can be expressed in the following block form:

[0093] Each block corresponds to the Laplace matrix of a fully connected subgraph in the original graph. Each block has an eigenvalue of 0 and has a unique solution, and the corresponding eigenvectors cover all nodes. Therefore, in the Laplace matrix of the original graph, the geometric multiplicity of the set of eigenvalue 0 is equal to the number of connected subgraphs, that is, the number of groups.

[0094] In order to improve the accuracy of ship risk group division and the structural consistency of grouping results, the present invention constructs a standard symmetric normalized Laplacian matrix ; further performs eigenvalue decomposition on it, sorts the eigenvalues of the Laplace matrix from small to large, and forms an n×k-dimensional eigenmatrix Q with the eigenvectors corresponding to the first k eigenvalues. Each row contains k samples as a sample. Using k-means clustering, the spectral clustering process implemented by this method can retain the global structure information of the original ship conflict network and effectively identify the high-risk density structure within the group on the basis of feature dimensionality reduction, and can obtain a clustering result with weak coupling between groups and strong risks within groups. In order to improve the practicality of the ship risk group division results in subsequent safety supervision and intelligent intervention, the present invention introduces a set of posterior risk assessment mechanisms after spectral clustering to quantify the local collision risk levels faced by each member ship in each group. Let P be a ship in the group, and n be the number of ships in the group that have a risk connection relationship with it, be the collision risk value between the th ship and P. Distinguish according to the collision danger level, and classify the risk relationship according to the set risk threshold: If , then it is the first type of data. At this time, it is considered that the risk between ship and P is relatively low; If , then it is the second type of data. At this time, the collision risk will have a certain impact on P; If , it is the third type, and the risk is relatively high at this time.

[0095] Through the above classification mechanism, the local interaction risk structure faced by each ship within the group is refined and evaluated, providing support for identifying potential high-risk individuals in the group, evaluating the coupling strength within the group, and formulating targeted supervision strategies.

[0096] In order to improve the ability to quantitatively express the overall risk level of a ship group, in the present invention, since the impacts on ships caused by different collision risk degrees are different, the comprehensive collision risk (Comprehensive Collision Risk Value, CCRV) of each ship is calculated. By statistically summing up the CCRV of each group, or by selecting the single ship with the highest CCRV within the group and comparing it with the group with the highest total CCRV or the highest single ship, the value with a higher ratio is taken as the risk value of the group. For the above-mentioned type I data, since the risk between the ship and P is considered relatively low, the impact on P can be temporarily ignored, and the output result is 0; for type II data, since attention is required, the maximum value is selected as the CCRV output result, that is:

[0097] where m is the number of type II data; for the above-mentioned type III data, since the risk is relatively high at this time and will have a significant impact on P, the output result is recorded as their sum, that is:

[0098] where n is the number of type III data. The sum of the CCRV of each ship within the group is the total comprehensive collision risk value, denoted by TCR, and the largest CCRV is denoted by MCR. The overall risk of a certain group is determined by TCR and MCR. Let the maximum value of TCR among all groups be max(TCR), and the maximum value of MCR be max(MCR). The degrees of collision risk caused by TCR and MCR in a certain group A are DT A and DM A , and the expressions are as follows:

[0099] The final degree of collision risk in the water area where each group is located depends on the higher value of DT and DM.

[0100] During a certain research period, the waters with the highest risk level continuously appeared on both the east and west sides of an island. This spatial distribution pattern mainly stems from geographical features and the characteristics of ship activities: the narrow waterway and the dense passenger ship flow form a compound risk source. In particular, the frequent intersection operations of passenger ferries and sightseeing ships significantly increase the collision probability. Therefore, it is urgent to establish a dynamic monitoring mechanism for this compound risk area to improve the ship collision avoidance ability and thus enhance the navigation safety factor. Through the analysis of temporal evolution, the collision risk in the southwest waters of the island reached its peak at 11:05. However, as the ship cluster in this area spreads and reorganizes, its risk potential value shows a gradient downward trend. It is worth noting that the risk level of the main waterway continued to rise during the period from 11:05 to 11:20 and jumped to the third risk hot spot area at 11:20. There is a significant spatial coupling between this high-risk zone and the navigation warning area delimited by the maritime department. This empirical correlation not only reveals the essential risk attributes of the waterway intersection nodes but also verifies the reliability of the method risk identification model in the present invention from the dimension of spatial matching.

[0101] As an implementation of the above method, the present invention provides a ship dynamic risk group identification system based on spectral clustering. The system includes the following modules: A data acquisition module configured to collect AIS data of coastal waters, decode, clean, repair, and interpolate the AIS data, and divide the processed data according to time windows to obtain preprocessed data; A data processing module configured to construct a ship encounter situation based on the preprocessed data and calculate the collision parameters between the own ship and the target ship. The collision parameters include: the relative distance between the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the shortest distance to CPA (DCPA) between the own ship and the target ship; A risk value calculation module configured to assign different weights to the shortest time TCPA and the shortest distance DCPA according to the ship encounter situation and calculate the collision risk value between ship pairs; A risk connection relationship graph construction module configured to construct a ship risk connection network graph with all ships as nodes and the risk values between ship pairs as the edge weights; A result output module configured to calculate the standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection network graph, extract multiple eigenvectors corresponding to the smallest eigenvalue of the Laplacian matrix, and arrange them in columns to form an eigenmatrix; Adopt the K-means clustering algorithm to perform clustering analysis on the eigenmatrix to obtain the division result of ship dynamic risk groups.

[0102] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the present application that have similar functions.

Claims

1. A method for identifying ship dynamic risk groups based on spectral clustering, characterized in that Including the following steps: S1. Collect AIS data of coastal waters, decode, clean, repair, and interpolate the AIS data, and divide the processed data by time window to obtain preprocessed data; S2. Based on the preprocessed data, construct a ship encounter situation, and calculate the collision parameters between the own ship and the target ship. The collision parameters include: the relative distance between the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the closest distance of approach (DCPA) between the own ship and the target ship; S3. According to the ship encounter situation, assign different weights to the shortest time TCPA and the shortest distance DCPA, and calculate the collision risk value between ship pairs; S4. Construct a ship risk connection relationship network graph with all ships as nodes and the risk values between ship pairs as the weights of the edges; S5. Based on the risk adjacency matrix of the ship risk connection relationship network graph, calculate the standard Laplacian matrix, and extract multiple eigenvectors corresponding to the smallest eigenvalue of the standard Laplacian matrix, and arrange them in columns to form an eigenmatrix; Use the K-means clustering algorithm to perform clustering analysis on the eigenmatrix to obtain the ship dynamic risk group division result.

2. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, wherein In step S3, according to the ship encounter situation, assign different weights to the shortest time TCPA and the shortest distance DCPA, and calculate the collision risk value between ship pairs, which specifically includes the following sub-steps: S31. Use the negative exponential function model to establish the relationship formula between the shortest time TCPA, the shortest distance DCPA, and the risk value to calculate the collision risk value of the ship in space and time. The specific calculation formula is: In the formula, represents the collision risk value based on the shortest distance DCPA; represents the collision risk value based on the shortest time TCPA; S32. According to different ship encounter situations, for , adopt different weight allocation strategies to calculate the collision risk value between ship and ship . The calculation formula is: In the formula, represents the comprehensive collision risk value between and ships; represents the weight coefficient under a specific ship encounter situation, and its value range is (0, 1); represents the weight complement corresponding to the shortest time TCPA; k represents the type number of the ship encounter situation, and its value is 1, 2, or 3, representing overtaking, head-on encounter, and crossing encounter respectively.

3. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, characterized in that, In step S4, construct a ship risk connection relationship network graph with all ships as nodes and the risk values between ship pairs as the weights of the edges, which specifically includes the following sub-steps: S41. Abstract all ships in the coastal waters as nodes of the relationship network graph; S42. According to the collision risk value between ship pairs calculated in step S3, establish connection edges between nodes, where: If the ship and the ship both have the minimum distance DCPA and the minimum encounter time TCPA lower than the preset safety threshold, then establish a connection edge; If either parameter of the shortest distance DCPA or the shortest time TCPA exceeds the preset safety threshold, it is regarded as a risk-free interaction and no connection edge is established; S43. Use the collision risk value as the weight of the edge to construct an undirected weighted graph; S44. Perform sparsification processing on the undirected weighted graph, retain the significant risk edges, and generate a ship risk connection relationship network graph.

4. The method for identifying dynamic risk groups of ships based on spectral clustering according to claim 1, wherein In step S5, before calculating the standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection relationship network graph, it also includes calculating the modularity of the ship risk connection relationship network graph to determine the optimal number of clustering clusters, which specifically includes the following sub-steps: S511, computing node degree of and node degree of ; S512. Calculate the total edge weight value m in the ship risk connection relationship network graph; S513. Calculate the expected edge weights of nodes and nodes in a random network . S514. According to the difference between the actual edge weight and the expected edge weight , the modularity M is calculated, and the calculation formula is: wherein, represents a node and node belong to the same group, otherwise it is 0; S515. Traverse different numbers of clustering clusters, and select the number of clusters that maximizes the modularity M as the number of ship groups used in spectral clustering.

5. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, characterized in that, In step S5, based on the risk adjacency matrix of the ship risk connection relationship network diagram, a standard symmetric normalized Laplacian matrix is constructed, and its calculation expression is: In the formula, I is the unit vector; D is the node degree matrix, which represents the number of nodes in the ship risk connection relationship network and is an n-order diagonal matrix, the diagonal elements are the degree values of the corresponding nodes, and the non-diagonal elements are all 0, where n is the number of ships; W is the risk adjacency matrix, and the risk adjacency matrix W is a symmetric matrix corresponding to an undirected graph, its diagonal elements are all 0, and the non-diagonal elements are 0 or 1.

6. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, wherein In step S5, it also includes posterior risk assessment of the clustered ship dynamic risk groups, which specifically includes the following sub-steps: S521. Let the target ship in a certain group be P, and traverse the collision risk values between other ships in the group and ship P , and classify them according to the following rules: If the collision risk value ∈[0, 0.4], the risk relationship is classified as type I data; If the collision risk value ∈(0.4, 0.8), the risk relationship is classified as type II data; If the collision risk value ∈ [0.8, 1], the risk relationship is classified as type III data; S522. According to the data categories divided in step S521, calculate the comprehensive collision risk value CCRV of each ship: For the first type of data, the comprehensive collision risk value CCRV is 0; For the second type of data, the comprehensive collision risk value CCRV is calculated through the following formula: In the formula, is the comprehensive collision risk value of the second type of data; m is the quantity of the second type of data; For the third type of data, the comprehensive collision risk value CCRV is calculated through the following formula: In the formula, is the comprehensive collision risk value of the type III data; n is the quantity of the type III data; S523. Calculate the total collision risk value TCR of the group according to the comprehensive collision risk value CCRV, and determine the maximum single-ship risk value MCR of the group. Obtain the first collision risk level value according to the total collision risk value TCR of the group , and obtain the second collision risk level value according to the maximum single-ship risk value MCR , and the expression is as follows: Wherein, is the first collision risk degree value of group A; is the total risk value of group A; is the maximum value among all group total risk values TCR; is the second collision risk degree value for Group A; is the maximum single-ship risk value for Group A; is the maximum single-ship risk value among all groups; S524. Determine the larger value between the first collision risk degree value and the second collision risk degree value as the final collision risk degree of the waters where Group A is located.

7. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, wherein In step S5, the K-means clustering algorithm is used to perform clustering analysis on the feature matrix, which specifically includes the following sub-steps: S531. Perform eigenvalue decomposition on the said standard Laplacian matrix and extract the eigenvectors corresponding to the first non-zero minimum eigenvalues; S532. Arrange the feature vectors by column to form a feature matrix Q of dimension , where is the number of ship nodes, and S533. Standardize each row of the feature matrix Q to obtain the standardized feature space data; S534. Use the K-means algorithm to cluster the feature space data, divide the ship dynamic risk groups, and output the group labels of each ship node.

8. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, characterized in that, In step S2, the calculation formulas for the shortest time to CPA TCPA when the own ship encounters the target ship and the shortest distance DCPA between the own ship and the target ship are: Wherein, OS and TS respectively represent the own ship and the target ship, d represents the relative distance, and respectively represent the speed of the own ship and the speed of the target ship, represents the relative speed, and β represent the relative bearing angle; The relative distance d is calculated based on the ship's longitude and latitude through the Haversine formula, and the Haversine calculation formula is: In the formula, represents a parameter; c represents the length of the major arc; represents the difference in latitude, represents the latitude; represents the difference in longitude, and R represents the radius of the earth; Relative speed is calculated based on the speed of the own ship and the speed of the target ship. The relative speed is calculated according to the following formula: In the formula, represents the course of the own ship, represents the course of the target ship.

9. The method for identifying ship dynamic risk groups based on spectral clustering according to claim 1, characterized in that In step S1, the AIS data is decoded, cleaned, repaired and interpolated, and the processed data is divided according to a time window to obtain preprocessed data, which specifically includes the following sub-steps: S11. Decode the AIS data and extract the position variables, speeds, routes, ship attributes and environmental parameters of the ships; S22. Clear the abnormal data, and the abnormal data includes longitude and latitude drift points, out-of-range data and incorrect MMSI identification codes; S23. Repair the missing trajectory data through cubic spline interpolation to generate a continuous trajectory sequence; S24. Divide the data at a preset time interval, construct a time window for dynamic risk analysis, and obtain preprocessed data.

10. A ship dynamic risk group identification system based on spectral clustering, characterized in that, The system includes: A data acquisition module, configured to collect AIS data in coastal waters, decode, clean, repair and interpolate the AIS data, and divide the processed data according to a time window to obtain preprocessed data; A data processing module, configured to construct a ship encounter situation based on the preprocessed data and calculate the collision parameters between the own ship and the target ship, where the collision parameters include: the relative distance between the own ship and the target ship, the relative bearing between the own ship and the target ship, the shortest time to CPA (TCPA) of the encounter between the own ship and the target ship, and the closest distance of approach (DCPA) between the own ship and the target ship; A risk value calculation module, configured to assign different weights to the shortest time to CPA (TCPA) and the closest distance of approach (DCPA) according to the ship encounter situation and calculate the collision risk value between ship pairs; A risk connection relationship graph construction module, configured to construct a ship risk connection network graph with all ships as nodes and the risk values between ship pairs as the weights of the edges; A result output module, configured to calculate a standard Laplacian matrix based on the risk adjacency matrix of the ship risk connection network graph, extract multiple eigenvectors corresponding to the smallest eigenvalue of the Laplacian matrix, and arrange them in columns to form an eigenmatrix; Use the K-means clustering algorithm to perform clustering analysis on the eigenmatrix to obtain the division result of the ship dynamic risk groups.

Images