Methods, devices, equipment and media for dividing ship traffic in complex waters

Abstract

The present invention discloses a method, device, equipment and medium for dividing ship traffic in complex waters, which divides regional traffic into conflict-related and spatially compact clusters to reduce the difficulty of understanding traffic patterns and risk management. First, a composite similarity measurement model that integrates conflict severity and spatial distance is designed, in which the conflict relationship between pairs of ships is quantified by a probabilistic conflict detection method, and the spatial compact relationship between ships is measured based on a maritime traffic route network constructed based on maritime knowledge learning. Subsequently, a clustering framework based on graph theory is used to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity. Finally, the AIS track data of Ningbo Zhoushan Port is used to comprehensively demonstrate and test the mechanism. Experimental results show that the mechanism 1) can decompose regional traffic well, 2) can identify high-risk / compact traffic clusters, and 3) is robust to various traffic scenarios in complex geographical waters.

Description

Technical Field

[0001] The present invention belongs to the technical field of transportation safety, and specifically relates to a method, device, equipment and medium for dividing ship traffic in complex waters. Background Art

[0002] Maritime traffic safety management plays a vital role in human life safety, economic development, and marine environmental protection. Active maritime surveillance based on modern intelligent transportation technologies is an indispensable component of maritime traffic safety management. To enhance maritime surveillance, a variety of systems have been developed and applied, such as vessel traffic services (VTS), automatic identification systems (AIS), land-based radar, and long-range identification and tracking (LRIT) systems. These systems have varying technical specifications and functions, enabling diverse aspects of maritime traffic monitoring and maritime situational awareness. However, the extensive surveillance areas and diverse vessel movements (such as navigation, berthing, anchoring, and refueling) pose significant potential challenges to their practical application. Especially in the context of economic globalization, increased traffic demand, and the emergence of larger ships, maritime traffic conditions have become more complex, hindering the effective implementation of these systems. Therefore, there is an urgent need for more advanced technologies and tools to help maritime operators better monitor and manage maritime traffic safety.

[0003] In response to the need to enhance monitoring in busy waters, various technologies and methods have been proposed for collision risk detection and assessment. These methods facilitate early collision warnings and support decision-making, providing a quantitative basis for maritime regulation. With the rapid development of AIS and the increasing availability of extensive ship motion information (i.e., AIS data), accurate collision risk assessment and maritime traffic characterization have become possible, further garnering widespread attention in recent years. In addition to maritime collision analysis, advanced applications of AIS data have been developed to address other problems, such as maritime traffic pattern extraction, maritime anomaly detection, port performance evaluation, and ship path optimization. While the application of AIS data facilitates accurate analysis and modeling of ship motion behavior, the complex traffic scenarios associated with multi-vessel interactions and the growing uncertainty in the spatiotemporal motion of ships present researchers with the challenge of developing more reliable models to support intelligent traffic situational awareness. More specifically, the assessment and perception of maritime traffic conditions requires continuous consideration of oceanographic features, the spatiotemporal dynamics and uncertainty of ship motion, and interference between multiple vessels. Therefore, developing effective traffic situational awareness models remains a challenging task, considering their practicality and applicability to complex traffic scenarios in busy waters. At the same time, how to adaptively divide the entire maritime traffic in a given busy water area into several sub-clusters to reduce the difficulty of situational awareness and further capture potential high-risk traffic clusters has remained largely unexplored. This suggests that further development of new advanced technologies is necessary to enhance the interpretation and understanding of traffic situations. Summary of the Invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a method, device, equipment and medium for dividing ship traffic in complex waters to solve at least one of the above-mentioned technical problems.

[0005] On the one hand, the present invention provides a method for dividing ship traffic in complex waters, including: constructing a composite similarity measurement model that integrates conflict severity and spatial distance; embedding the similarity measurement results into a graph theory clustering method, and using a graph theory-based clustering framework to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity.

[0006] The above technical solution takes into account that the conflict severity metric incorporates the uncertainty characteristics of ship movement in dynamic waters, while the spatial compactness metric extracts the shortest path between two ships using the maritime traffic route network mined from maritime trajectory knowledge, and integrates these two indicators into a composite similarity metric model. The similarity metric results are then further embedded in the graph theory clustering method to generate traffic clusters that ensure maximum similarity within the cluster, minimum similarity between clusters, and balanced size, thereby achieving the goal of dividing regional maritime traffic into several compact, scalable and interpretable traffic clusters.

[0007] As a further technical solution, the method further includes: using a kernel density estimation algorithm to obtain the probability distribution results of navigable areas and non-navigable areas; based on the probability distribution results, converting the entire water area into a binary image, using image processing technology to extract the image skeleton, and constructing a ship traffic route network; using the ship traffic route network to identify the real spatial distance between pairs of ships.

[0008] In complex and restricted waters, two spatially adjacent ships may not be able to reach each other. The traditional linear distance metric is not suitable for describing the spatial compactness of traffic scenes in waters with restricted geographical features. Therefore, a ship traffic route network is constructed based on traffic knowledge extraction, and then the shortest distance between ship pairs is searched on the route network as their actual spatial distance.

[0009] As a further technical solution, the real spatial distance between the pairs of ships is identified using the ship traffic route network, further including: uniformly sampling a number of points on the connecting line between the positions of the two ships, and determining whether they belong to the navigable area; if all these points are located in the navigable area, the real spatial distance between the pairs of ships is calculated based on the Euclidean distance.

[0010] As a further technical solution, as long as there is a point in the unnavigable area, the two adjacent points closest to each ship are searched on the traffic route network, and then the Dijkstra algorithm is applied to calculate the shortest path distance between the two points.

[0011] As a further technical solution, the method combines the conflict connection similarity metric and the spatial compactness similarity metric by a linear combination method; wherein the conflict connection similarity metric is defined as equal to the conflict severity between ships, and the spatial compactness similarity metric is defined as:

[0012]

[0013] Among them, dist ij represents the real space distance between the two ships, β is a scaling parameter, and D1 and D2 are two parameters that divide the spatial compactness relationship into three categories.

[0014] As a further technical solution, the constructed composite similarity measurement model is:

[0015]

[0016] Among them, W ij Defines the similarity between the two ships, α is a trade-off parameter, represents a spatially compact similarity metric, Represents the conflict connection similarity measure.

[0017] As a further technical solution, the method further comprises:

[0018] The Ncut objective function is used to generate clusters with high internal similarity and low external similarity:

[0019]

[0020] in, W(u, v) represents the similarity between samples u and v, A i represents the subset of the i-th cluster, Representative A i The complement of

[0021] By using the normalized similarity matrix Alternative The above objective function is transformed into the Frobenius norm minimization form;

[0022] Given a normalized similarity matrix The number of clusters k, non-negativity constraint Based on SNMF graph clustering optimization problem is transformed into the following form:

[0023]

[0024] The purpose of SNMF is to search for a symmetric non-negative low-order approximation H by minimizing the Frobenius norm in the above formula, and directly identify the cluster assignment of data samples according to the maximum value in each row of H.

[0025] On the one hand, the present invention provides a device for dividing ship traffic in complex waters, including: a similarity measurement construction module, used to construct a composite similarity measurement model that integrates conflict severity and spatial distance; a traffic cluster division module, used to embed the similarity measurement results into a graph theory clustering method, and use a graph theory-based clustering framework to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity.

[0026] Furthermore, the similarity measurement construction module also includes: a conflict connection similarity measurement module and a space compactness similarity measurement module. Through the linear combination of these two measurement indicators, a composite similarity measurement result is obtained.

[0027] On the one hand, the present invention provides a computer device, which includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method for dividing ship traffic in complex waters are implemented.

[0028] On the one hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method for dividing ship traffic in complex waters are implemented.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] (1) The present invention considers the multi-attribute relationships among multiple ships in the traffic segmentation process (i.e., conflict connectivity and spatial compactness). Compared with other traffic cluster detection methods that only focus on traffic density, this method is more reliable and practical because it not only helps to identify traffic conflict patterns but also makes the application of traffic management strategies feasible.

[0031] (2) The present invention utilizes historical AIS data to generate a maritime traffic route network. It helps capture traffic clusters with real spatial compactness by using the shortest path lengths of ship pairs on the route network, ensuring adaptation to traffic scenarios in restricted geographical waters.

[0032] (3) This paper applies an extended graph-based clustering framework, namely symmetric non-negative matrix factorization (SNMF), to generate ideal traffic clusters with balanced sizes. By optimizing the graph-based clustering objective, it can flexibly and scalably handle various ship traffic encounter scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flowchart of a method for dividing ship traffic in complex waters according to an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram of unnavigable area recognition based on kernel density estimation and ship traffic route network extraction based on image processing technology according to an embodiment of the present invention.

[0035] Figure 3 Schematic diagram of actual spatial distance calculation based on a traffic route network according to an embodiment of the present invention.

[0036] Figure 4 Schematic diagram of the ship traffic division results according to an embodiment of the present invention, where (a) represents the visualization of the ship traffic network; (b) represents f1 and f2 corresponding to different numbers of clusters; (c)-(f) represent the corresponding traffic cluster division results when the number of clusters is 11, 14, 17, and 20.

[0037] Figure 5 (a) is a method according to an embodiment of the present invention Figure 4 (d) Schematic diagram of the feature statistics of each cluster.

[0038] Figure 5 (b) is a method according to an embodiment of the present invention Figure 4 (f) Schematic diagram of the feature statistics of each cluster.

[0039] Figure 6 (a)-(b) are schematic diagrams showing the performance comparison between the algorithm proposed according to an embodiment of the present invention and the spectral clustering algorithm under different numbers of ships and clusters. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions of various embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0041] The present invention aims to divide regional maritime traffic into several compact, scalable and interpretable traffic clusters. The biggest challenge in dividing regional ship traffic is to explain its pattern complexity and mutual interference between ships. Inspired by the work on urban traffic network division in the field of road traffic, a graph-theory-based traffic division mechanism is designed to generate traffic clusters that ensure conflict connectivity and spatial compactness. The conflict severity metric incorporates the uncertainty characteristics of ship motion in dynamic waters, while the spatial compactness metric extracts the shortest path between two ships using the maritime traffic route network mined from maritime trajectory knowledge. These two indicators are integrated into a composite similarity measurement model through linear combination. Finally, the similarity measurement results are further embedded in the graph theory clustering method to generate traffic clusters that ensure the maximum similarity within the cluster, the minimum similarity between clusters, and the balanced size. In addition, some indicators are used to evaluate and test the performance of the proposed traffic division method.

[0042] The primary objective of this invention is to reduce the difficulty of regional traffic risk perception by capturing real-time high-risk traffic clusters in complex, confined waters. More specifically, we seek to develop a mechanism to partition the entire vessel traffic into clusters to achieve the following goals: 1) extract traffic clusters with high conflict relationships to detect true traffic conflict patterns; and 2) generate spatially compact traffic clusters to facilitate the design and implementation of traffic management strategies.

[0043] Based on these two goals, the present invention constructs an undirected graph for ship traffic partitioning, in which each ship is modeled as a node, and its adjacent relationship (i.e., edge) is established based on the conflict relationship and spatial distance between ships. Doing so can transform the traffic partitioning problem into a graph cutting problem. It is committed to partitioning the network into several subgraphs. The proposed traffic partitioning mechanism includes the following main steps. First, a composite similarity model that simultaneously considers conflict connectivity and spatial compactness is proposed. Among them, a probabilistic conflict detection method is used to quantify the conflict relationship, which can accurately estimate the comprehensive conflict severity between ships in the presence of uncertainty in the spatiotemporal motion of ships. On the other hand, the spatial compact relationship between ships is measured based on the maritime traffic knowledge extraction technology. It extracts the real spatial distance between ship pairs from the constructed ship traffic route network. On the basis of the constructed composite similarity model, the mathematical framework of graph clustering is further used to group ships with high conflict connectivity and spatial compactness into clusters of balanced size. In addition, some indicators are used to evaluate and test the performance of the proposed traffic partitioning framework. The specific process is as follows Figure 1 shown.

[0044] A key issue in graph partitioning is how to define similarity / adjacency to describe the connection between each pair of nodes. This paper focuses on developing a similarity model that simultaneously considers the conflict relationships and spatial distances between pairs of ships. The similarity model includes the following modules: 1) a probabilistic conflict assessment model that reflects conflict relationships, 2) a proposed true spatial distance recognition model to define spatial relationships, and 3) a composite similarity model that describes the design.

[0045] Collision risk quantification is an integral part of conflicting traffic cluster identification. Here, the conflict relationships between ships are measured probabilistically to ensure adaptation to traffic scenarios with spatiotemporal motion uncertainties.

[0046] Typically, a conflict is defined as a violation of the minimum safe separation distance between two ships within a limited prediction timeframe. In this paper, a classic ship domain model suitable for confined waters is used to describe conflicts between ships. A conflict is considered to have occurred if the following inequality holds within the conflict detection timeframe:

[0047] Dist AB (t)≤SD A (t)+SD B (t) (1)

[0048] Among them, Dist AB Indicates the distance between the two ships, SD A and SD Brepresents the distance of each ship to the boundary of their territory. Considering the impact of ship motion uncertainty on conflict detection, whether inequality (1) holds is a probabilistic event. Therefore, the probability of conflict between ships at a given time t is expressed by the following expression:

[0049]

[0050] Among them, f L(t) The probability density function of the loss of the minimum safe distance between two ships is L(t) = Dist(t) - SD A (t)-SD B (t).

[0051] Equation (2) represents the instantaneous probability of a conflict occurring at a certain time in the future. Therefore, the severity of the conflict is expressed by considering the maximum PC(t) and its corresponding occurrence time. The first indicator reflects the highest intensity of the potential conflict, while the second reflects the urgency of resolving the conflict immediately. Here, an exponential function is used for fusion, as shown below:

[0052]

[0053] Where MPC represents the maximum probability of conflict within the detection time range, t MPC Indicates the time when the maximum conflict probability occurs.

[0054] Defining a conflict severity measurement model from a probabilistic risk perspective allows for probabilistic conflict detection. Probabilistic conflict detection methods include two key elements: uncertain trajectory prediction, a prerequisite for potential collision detection and assessment; and conflict probability calculation, which involves the development and implementation of techniques and methods that accurately and efficiently estimate the severity of actual conflicts. Uncertain trajectory prediction is determined by modeling ship motion as a deterministic motion related to the ship's navigation plan, combined with a random component introduced by various disturbances. For conflict probability calculation, a two-stage Monte Carlo simulation algorithm is employed to rapidly and accurately estimate conflict severity.

[0055] Maritime traffic segmentation requires ensuring the spatial compactness of the generated traffic clusters to facilitate the implementation of collision risk management strategies. The traditional spatial compactness metric is determined based on the Euclidean distance (or linear distance) between ships in a cluster. However, in complex and confined waters, two spatially adjacent ships may not be able to reach each other. For example, there are often obstacles between ships that block them. Therefore, the traditional linear distance metric is not suitable for describing the spatial compactness of traffic scenarios in waters with restricted geographical features.

[0056] An effective approach to addressing this problem is to construct a ship traffic route network based on traffic knowledge extraction, and then search for the shortest distance between pairs of ships on the route network as their actual spatial distance. This invention proposes to address this problem by employing an image preprocessing technique. It uses the main framework of navigable waters as the traffic route network.

[0057] First, kernel density estimation is applied to distinguish navigable waters from non-navigable waters. It estimates the spatial probability distribution of ship traffic based on AIS data using the following formula:

[0058]

[0059] Among them, φ h Indicates that φ is satisfied h (x)>0 and ∫φ h The kernel function (x)dx = 1 is used, h represents a bandwidth parameter greater than 0, and K' represents the number of samples within the bandwidth h. For each area point, if its probability distribution value is greater than a predefined threshold, it indicates a navigable area; otherwise, it is considered unnavigable waters.

[0060] Using the probability distribution obtained from kernel density, the entire water area can be converted into a binary image, where cells 1 represent navigable areas and 0 represent non-navigable areas. Image processing operations are applied to the binary image to extract the image skeleton. Through a series of operations, a network framework is constructed that provides a compact, structured, and accurate description of traffic routes.

[0061] After completing the offline training of the maritime traffic route network structure, it can be further used to identify the real spatial distance between pairs of ships. The program implementation includes the following steps. First, several points (such as 10) are evenly sampled on the connecting line between the positions of two ships, and it is determined whether they belong to the navigable area. If all these points are located in the navigable area, the actual spatial distance between the pairs of ships is calculated based on the Euclidean distance. Otherwise, the nearest point adjacent to each ship is searched on the traffic route network, and then the Dijkstra algorithm is applied to calculate the shortest path distance between the two points. In this way, the program provides the potential to support the generation of real spatial compact traffic clusters.

[0062] Based on the above two relationship metrics, they can be further merged to simultaneously meet the requirements of spatial compactness and conflict connectivity. In the present invention, the two metrics are combined together by a linear combination method. This is a simple and powerful method to describe the association relationship between pairs of ships, but requires that the two metrics have the same value range. Since the value range of the conflict severity between ships is [0, 1], the similarity metric of the conflict connection relationship between ships is defined as equal to its conflict severity. However, the actual spatial distance between ships may be very far (for example, tens of nautical miles). Therefore, the spatial compact relationship similarity metric is as follows to ensure that its value is between 0-1:

[0063]

[0064] Among them, dist ij represents the real spatial distance between the two ships, β is a scaling parameter, and D1 and D2 are two parameters that divide the spatial compactness relationship into three categories. If the spatial distance is less than D1, the two ships are considered to be highly compact and If the spatial distance is between D1 and D2, the compact similarity decreases monotonically with the spatial distance based on an exponential expression; if the spatial distance is greater than D2, the spatial compactness relationship between the two ships can be ignored.

[0065] Furthermore, a composite similarity measurement model is designed and given by linear combination. and Assign different weights as follows:

[0066]

[0067] Among them, W ij The similarity between two ships is defined, and α is a trade-off parameter. The model explicitly considers the two similarity measures mentioned above, which helps to systematically describe the multiple mutual relationships between ships in the entire water area.

[0068] Spectral clustering is a class of clustering algorithms that are widely used to solve graph partitioning problems. Unlike other classes of clustering algorithms that focus on the dataset itself (e.g., prototype-based and density-based clustering), spectral clustering assigns data samples to appropriate clusters based on the similarity between each pair of data samples and does not make assumptions about the form of the clustered dataset. However, the successful application of spectral clustering depends heavily on the properties of the main eigenvalues ​​and eigenvectors of the Laplacian matrix (L = DW), where W represents the similarity matrix and D defines a diagonal matrix When the gap between the kth and (k+1)th largest eigenvalues ​​of the matrix L is not large enough, the application of spectral clustering may fail due to the instability of the k-dimensional subspace scanned by the first k eigenvectors of L.

[0069] To solve this problem, a more competitive extended mathematical formula, namely SNMF, is proposed. It distinguishes different clusters by performing a non-negative low-rank approximation on the graph similarity matrix. In theory, according to the graph clustering goal, SNMF and spectral clustering are two highly related methods, but they use fundamentally different methods to optimize the goal. The present invention adopts SNMF. In order to divide marine ship traffic into clusters of similar size, a common objective function called normalized cut (Ncut) is adopted to generate clusters with high internal similarity and low external similarity, which is defined as follows:

[0070]

[0071] in, W(u, v) represents the similarity between samples u and v, A i represents the subset of the i-th cluster, Representative A i By using the normalized similarity matrix Alternative Formula (7) can be transformed into the Frobenius norm minimization form. Therefore, given the normalized similarity matrix The number of clusters k, non-negativity constraint The SNMF graph clustering optimization problem can be transformed into the following form:

[0072]

[0073] The purpose of SNMF is to search for a symmetric non-negative low-order approximation H by minimizing the Frobenius norm in formula (8). For the optimal matrix H, each column can be regarded as the membership of a cluster sample to a cluster. Therefore, the cluster assignment of the data sample can be directly identified based on the maximum value in each row in H. Note that different optimization methods can be considered to solve the minimization problem described in equation (8). In the present invention, the optimization problem is directly solved by a Newton-like algorithm. Due to its sensitivity to the initialization of H, it may encounter a local minimum solution. Regarding this problem, the Newton-like algorithm is executed multiple times (for example, 10 times) with a randomly sampled initialization H to find the global minimum solution or at least ensure that it is close to the global minimum.

[0074] Model performance evaluation is crucial to ensure the effectiveness of the proposed method. Therefore, some metrics are introduced to evaluate the traffic segmentation results.

[0075] The first metric used was the "NCUT Profile" (NS), which is represented as follows:

[0076]

[0077] Among them, A i and Aj Represents two clusters, k represents the number of clusters, W(u, v) represents the similarity between two samples, Indicates A i The number of samples in NS k (A i , A j ) Calculate cluster A i and A j On this basis, the following indicators are used to measure whether the ships in a cluster are correctly grouped:

[0078]

[0079] Among them, NSN k (A i , A j )=min{NS k (A i , A K )|A K ∈A,A K ≠A i}, A represents the entire cluster set, A j Defined as i The most similar cluster. Obviously, NS k (A i )<1 indicates cluster A i Furthermore, the average NS of all clusters can be used to K To evaluate the overall performance of the traffic segmentation results, as follows:

[0080]

[0081] Smaller NS K A value of means that the entire traffic scene is effectively segmented.

[0082] It is worth noting that the above comprehensive metrics are highly dependent on the designed similarity model. Due to the influence of hyperparameters in the similarity model, they cannot directly examine the spatial compactness and conflict connectivity of the cluster partitioning results. Therefore, two specific metrics related to these two criteria are further proposed. Regarding conflict connectivity, it can be evaluated based on the degree to which conflicting ship pairs are divided into different clusters, as shown below:

[0083]

[0084] Among them, N vc represents the number of conflicting ship pairs assigned to different clusters, C(γ) i represents the severity of the conflict between ship pair i. A smaller f1 value indicates that more conflicting ship pairs are effectively assigned to the same cluster. Regarding spatial compactness, NSk It can still be applied by replacing the dissimilarity in Equation (9) with the actual spatial distance, which is called f2. The smaller the f2 value, the smaller the spatial distance within the cluster and the larger the spatial distance between clusters, which means that the spatial compactness of the cluster partition is well achieved.

[0085] In this embodiment, Figure 2 The figure shows a network of unnavigable areas and vessel traffic routes, identified through knowledge mining of historical AIS data. The gray areas represent unnavigable areas, while the curve represents the traffic route network. The study found that the curve effectively describes the skeleton of the navigable area, demonstrating its goodness of fit. More importantly, this precise and structured representation of ocean behavior patterns allows for the measurement of real-world spatial distances between pairs of ships. To illustrate how the network can be used for real-world spatial distance calculations, Figure 3 An example of identifying the spatial distance relationship between pairs of ships based on the created topology is given. In the figure, ships B and C are separated from A by an obstacle. The real spatial distance (RSD) and linear spatial distance (LSD) between ships A and B based on the transportation network are 6.43 kn and 4.51 kn, respectively. Obviously, due to the presence of obstacles between the pairs of ships, it is more reasonable to measure the distance between the pairs of ships by their shortest path length on the route network rather than using physical distance. In other words, the route network helps to identify the real spatial distance in complex waters, which is the first step to identifying real spatial compact traffic clusters.

[0086] Figure 4 The ship traffic division results of the traffic scene at a certain moment are given. Figure 4 A visualization of the graphical representation of ship traffic relationships is shown in (a). Red dots represent ships, thin lines represent similarities between pairs of ships, and thick lines represent conflict relationships between pairs of ships. Figure 4 (b) shows the values ​​of f1 and f2 after traffic segmentation is performed using different numbers of clusters. Figure 4 The multiple troughs of the dashed lines in (b) show several traffic partitioning results (see Figure 4 (cf)). It is obvious from these figures that the generated traffic clusters are spatially compact, while at the same time most conflicting ship pairs are assigned to the same cluster, which illustrates the good properties of our model. In addition, it is found that there are complex conflict relationships between ships (e.g. Figure 4Clusters 2 and 3 in (e) indicate that the spatial and temporal interactions of multiple ships should be considered, rather than focusing on interactions between pairs of ships. Furthermore, the clustering quality is robust to varying numbers of input clusters, and as the number of clusters increases, more outliers are filtered out (i.e., a cluster with a single ship is generated, which can be considered a safe vessel). This means that one can choose different numbers of clusters to study the clustering results of traffic scenarios for multi-perspective analysis. Overall, the proposed method performs well in partitioning regional ship traffic into several highly spatially compact and conflict-connected clusters.

[0087] also, Figure 4 The characteristics of the clusters generated in (d) and (f) are analyzed. Clusters with less than 3 ships are not marked here. Figure 5 The sum of traffic density and conflict severity for each cluster is shown. As shown in these figures, clusters with high density / conflict severity can be easily found, e.g. Figure 5 Cluster 3 in (a) and Figure 5 Cluster 9 in (b). This demonstrates the need to decompose regional traffic complexity rather than directly implementing situational awareness from a global perspective. Regarding the practical application of the proposed method, it is possible to check whether the risk index of a traffic cluster exceeds a predefined threshold, helping monitoring operators focus on critical traffic clusters. In this way, the proposed traffic decomposition mechanism helps reduce the difficulty of situational awareness and identify high-risk / density areas.

[0088] Figure 4 Feature statistics of each cluster in (d) and (f), including the sum of traffic density and conflict severity.

[0089] Model comparison and validation are crucial for the practical application of modeling methods. Therefore, we first compare the proposed method with a widely used graph clustering algorithm (i.e., spectral clustering) to show the superiority of the SNMF framework.

[0090] Table 1. Comprehensive comparison between SNMF and spectral clustering

[0091]

[0092] Table 1 gives a comprehensive comparison between the proposed mechanism and the spectral clustering algorithm. As shown in the table, the overall performance of the proposed method is significantly better than spectral clustering in terms of multiple evaluation indicators. This is mainly because the good properties of the SNMF framework and the orthogonality constraint have no effect on it. Please note that other categories of clustering algorithms (such as prototype-based and density-based clustering) are not considered for comparison because they focus on the characteristics of each data sample. For example, the k-means algorithm clusters based on cluster centers, which is meaningless when the spatial distance between ships is measured by their shortest path length on the route network instead of the Euclidean distance. The DBSCAN algorithm requires the identification of core samples and has difficulty handling data sets with different densities. Therefore, it is difficult for these models to achieve effective traffic clustering based on the interactions / similarity between ships. In order to further evaluate the generalization ability of the proposed mechanism, the two methods are compared under different numbers of clusters and numbers of ships. Figure 6 As shown in Figure 3, the proposed method outperforms the spectral clustering algorithm based on average NS in all cases. These results confirm the stability and scalability of the SNMF framework.

[0093] The development of advanced traffic situational awareness technologies and tools is one of the important components of emerging smart ports. In this paper, a graph-based traffic partitioning mechanism is proposed, which allows the capture of conflicting connections and spatially compact traffic clusters to reduce the difficulty of global situational awareness and reduce the difficulty of designing and implementing traffic management strategies. The mechanism has the following characteristics: 1) It takes into account the multi-attribute relationships between ships, including conflict relationships and spatial distances; 2) The method extracts and identifies accurate spatial distances based on maritime traffic knowledge, enabling the mechanism to adapt to complex geographical waters; 3) The study adopts a more competitive graph clustering model to support robust traffic cluster partitioning. The experiment conducted a correlation analysis based on AIS data to verify the practicality and superiority of the study. The experimental results show that the method is of great significance in supporting intelligent maritime monitoring and promoting maritime traffic safety management.

[0094] This embodiment provides a device for dividing ship traffic in complex waters, including: a similarity measurement construction module for constructing a composite similarity measurement model that integrates conflict severity and spatial distance; a traffic cluster division module for embedding the similarity measurement results into a graph theory clustering method, and using a graph theory-based clustering framework to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity.

[0095] Furthermore, the similarity measurement construction module also includes: a conflict connection similarity measurement module and a space compactness similarity measurement module. Through the linear combination of these two measurement indicators, a composite similarity measurement result is obtained.

[0096] The apparatus provided in the above embodiment may be implemented in the form of a computer program, which may be run on a computer device.

[0097] The computer device may include a processor, a memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to execute any of the complex waters vessel traffic demarcation methods.

[0098] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0099] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any method for dividing ship traffic in complex waters.

[0100] The network interface is used for network communication, such as sending assigned tasks, etc. It will be understood by those skilled in the art that the aforementioned structure is merely an illustration of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0101] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] In one embodiment, the processor is configured to execute a computer program stored in the memory to implement the following steps:

[0103] Construct a composite similarity measurement model that integrates conflict severity and spatial distance;

[0104] The similarity measurement results are embedded in the graph theory clustering method, and the graph theory-based clustering framework is used to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity.

[0105] This embodiment further provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of this application.

[0106] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMediaCard (SMC), a Secure Digital (SD) card, a flash memory card, etc., equipped on the computer device.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for dividing ship traffic in complex waters, characterized by: include: A composite similarity measurement model that integrates conflict severity and spatial distance is constructed. The similarity measurement results are embedded in a graph-theoretic clustering method, and a graph-theoretic clustering framework is used to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity. The method further comprises: obtaining a probability distribution result of navigable areas and non-navigable areas using a kernel density estimation algorithm; converting the entire water area into a binary image based on the probability distribution result, extracting the image skeleton using image processing technology, and constructing a ship traffic route network; and identifying the real spatial distance between pairs of ships using the ship traffic route network; The method combines two metrics, the conflict connection similarity metric and the spatial compactness similarity metric, through a linear combination method. The conflict connection similarity metric is defined as equal to the conflict severity between ships, and the spatial compactness similarity metric is defined as: ， in, represents the real space distance between the two ships, is a scaling parameter, and are two parameters that divide spatial compactness relations into three categories; The constructed composite similarity measurement model is: ， in, The similarity between the two ships is defined. is a trade-off parameter. represents a spatially compact similarity metric, represents the conflict connection similarity measure; The method further comprises: The Ncut objective function is used to generate clusters with high internal similarity and low external similarity: ， in, , Representative samples u and v The similarity between Representative i A subset of clusters, represent The complement of By using the normalized similarity matrix Alternative , the above objective function is transformed into the Frobenius norm minimization form; Given a normalized similarity matrix , the number of clusters k , non-negativity constraint , based on SNMF graph clustering optimization problem is transformed into the following form: ， The purpose of SNMF is to search for symmetric non-negative low-order approximations by minimizing the Frobenius norm in the above formula. H ,according to H The maximum value in each row of is used to directly identify the cluster assignment of the data sample.

2. The method for dividing ship traffic in complex waters according to claim 1 is characterized in that: Identifying the true spatial distance between pairs of ships using a ship traffic route network further includes: uniformly sampling a number of points on a connecting line between the positions of two ships, and determining whether they belong to a navigable area; if all of the points are located in the navigable area, calculating the true spatial distance between the pairs of ships based on the Euclidean distance.

3. The method for dividing ship traffic in complex waters according to claim 2 is characterized in that: As long as there is a point in the unnavigable area, the two adjacent points closest to each ship are searched on the traffic route network, and then the Dijkstra algorithm is applied to calculate the shortest path distance between the two points.

4. The device for dividing ship traffic in complex waters is characterized by: include: A similarity measurement construction module is used to construct a composite similarity measurement model that integrates conflict severity and spatial distance; a kernel density estimation algorithm is used to obtain the probability distribution results of navigable and non-navigable areas; Based on the probability distribution results, the entire water area is converted into a binary image, and the image skeleton is extracted using image processing technology to construct a ship traffic route network. The ship traffic route network is used to identify the real spatial distance between pairs of ships. The conflict connection similarity measure and the spatial compactness similarity measure are combined by a linear combination method. The conflict connection similarity measure is defined as equal to the conflict severity between ships, and the spatial compactness similarity measure is defined as: ， in, represents the real space distance between the two ships, is a scaling parameter, and are two parameters that divide spatial compactness relations into three categories; The constructed composite similarity measurement model is: ， in, The similarity between the two ships is defined. is a trade-off parameter. represents a spatially compact similarity metric, represents the conflict connection similarity measure; The traffic clustering module is used to embed the similarity measurement results into the graph theory clustering method, and use the graph theory-based clustering framework to generate balanced traffic clusters with high intra-cluster similarity and low inter-cluster similarity; the Ncut objective function is used to generate clusters with high internal similarity and low external similarity: ， in, , Representative samples u and v The similarity between Representative i A subset of clusters, represent The complement of By using the normalized similarity matrix Alternative , the above objective function is transformed into the Frobenius norm minimization form; Given a normalized similarity matrix , the number of clusters k , non-negativity constraint , based on SNMF graph clustering optimization problem is transformed into the following form: ， The purpose of SNMF is to search for symmetric non-negative low-order approximations by minimizing the Frobenius norm in the above formula. H ,according to H The maximum value in each row of is used to directly identify the cluster assignment of the data sample.

5. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the complex waters vessel traffic division method as described in any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the complex waters vessel traffic division method according to any one of claims 1 to 4 are implemented.

Images